A SUMMARIZED REPORT ON TCP PERFORMANCE ENHANCEMENT ON DATA TRANSFER USING SCHEDULING ALGORITHM Research paper by: A.Bharathi, K. Ananda Kumar, A. Shanumugam
Summarized by: Mahan Malik MCA Vth Sem 0712814023
ABSTRACT This paper analyzes the TCP performance for both bulk transfer and small packet transfer traffic separately to reveal some of the issues and the possible solution in the TCP performance. This paper develops an efficient scheduling algorithm for TCP bulk data transfer traffic, which set up a TCP receive window that is efficiently and to make the data movement as efficient as possible. Request reply types of application typically generate small traffic .the latency resulting from the packet is also more important as that of the throughput delivered .in this paper, the developed scheduling algorithm also provides a technique to reduce the latency time.
INTRODUCTION TCP is having following characteristics:1) Unicast Protocol 2) Connection State 3) Reliable 4) Full Duplex 5) Streaming 6) Rate Adaptation
LITERATURE REVIEW Categorizing TCP traffic:TCP is basically having two type of transfer traffic Bulk Transfer Traffic-the payload size of most segments from sender to receiver is 1460 bytes(Ethernet is having maximum frame size of 1518 bytes. it typically contains 14 bytes fro header and 4 bytes for CRC.then from the remaining 1500 bytes 20 bytes of IP header and 20 bytes of TCP header are reduced which resulted remaining 1460 bytes). For example, FTP transfers and downloading web pages with large amount of graphics. Small Packet Traffic-the payload size of most segments from sender to receiver is below 1460 bytes. For e.g. A request from the clients and a short reply by the server. Bottlenecks:A high speed line sends small messages so quickly that it creates a massive amount of interrupt pressure within the kernel space of the receiving host. Furthermore the kernel must deal with processing packets and moving messages payloads into user space after receipt and error checking .High communication overhead can overwhelm the processor and prevents it from spending valuable time on computation. This processing time also
prevents messages from being delivered to the application quickly. Due to these blockades, application cannot harness the bandwidth and speed the network provides. Transmission Latency:Many high performances, scientific computing applications depend on rapid, low latency transmission of messages between processors. High message latency leads to CPU idling and wasting resources. The application may wait for message arrival before continuing computation. Ensuring consistently low messages latency processes frequently sending messages between nodes suffer performance loss in a high latency environment. PROBLEM DEFINITION There are two problems that have been observed. One is Small Packet Problem: - this problem occurs when we transfer small data packets. Suppose we want to send 1 byte of data then we have to attach 40 byte of header (20 byte of IP header and 20 byte of TCP header), then this will increase the overhead which can result in congestion, lost datagram and retransmission. In practice, may drop so low that TCP connections are aborted. Bulk Data Transfer Traffic: -This problem occurs when the amount of data to move from one computer to another is far larger like, FTP transfer, heavy graphics downloading from web etc. PHASES Small packet transfer and scheduling algorithm: - To reduce transmission latency during small packet transfer we use Inter user scheduling priority model by deploying fair queuing with strict priority or rate priority. Each user reports measured condition to the pf scheduler. User with best channel is selected to transmit in different time slot.PF weight the current rate achievable by average rate received by user and select user. Bulk Data transfer and scheduling algorithm: - The objective of this phase is to maximize the efficiency of data transfer , implying that TCP should endeavor to locate the maximum point dynamic equilibrium of maximum network efficiency , where the sending data rate is maximized just prior to onsets of packet sustained loss. Further increasing the rate from such a point will run the risk of generating a congestion condition within the network, with rapidly increasing packet loss levels. METHODOLOGY Size of TCP window: -The size of the window can be determined by the given formula. (X*T)/8 where=rate in bps T=latency time. To get one gigabit per second on Ethernet, the system must deliver 1,000,000,000/8/1518=82,345 packets per second. This is equivalent to delivering one full sized packet every microsecond. If the latency is 100 microseconds the size of the window needs to be at least 1,000,000,000*0.0001/8=12,500 bytes,
Overhead to move data from the sender to the receiver: - The network latency is calculated as measurement time over the number of packets transferred hence if the measurement time is 1 sec. latency is the reciprocal of packet rate per sec. Since the packet rate also shows the server’s capability to process packets besides latency, this paper uses this metric for the processing of data on small packet traffic. PERFORMANCE EVALUATION AND RESULTS: - The ns-2 simulator was used to evaluate the proposed mechanism. In order to determine how well the proposed algo performs under various conditions, three different scenarios were generated by considering the following factors: • Throughput • Dataloss • Delay Case -1 For larger packets No. of packet Sent
No. of packet received
1270
793
1670
918
2070
1168
2470
1168
2.Data Loss No. of packet Sent
Data loss
1270
477
1670
752
2070
1027
2470
1302
3.Delay No. of packet Sent
Delay
1270
10
1670
12
2070
14
2470
16
For small packet Transfer:-1.Throughput No. of packet Sent
2.Dataloss
No. of packet received
1270
773
1670
898
2070
1023
2470
1148
No. of packet Sent
Data loss
1270
477
1670
752
2070
1027
2470
1577
3.Delay No. of packet Sent
Delay
1270
10
1670
12
2070
14
2470
16
Conclusion:1.By developing an efficient algorithm for small packet and bulk data transfer we could be able to reduce enough overhead from TCP/IP. 2.By appropriately offloading small parts of protocols functionality on to a NIC, we can reduce latency and optimized bandwidth utilization for distributed computing environment REFERENCES 1.Marco melia, michela Meo and Claudo Casetti,”TCP Smart Framing:ASegmentation Algorithm To reduce Latency”.IEEE/ACM transaction on networking.
2.Internet options 3.Karpagam JCS VOL.3 Issue 2 Jan-Feb 09