Congestion Control

FuzzyCCG: A Fuzzy Logic QoS Approach for Congestion Control in Wireless Ad hoc Networks Lyes Khoukhi

Soumaya Cherkaoui

Department of Electrical and Computer Engineering, Université de Sherbrooke J1K 2R1, QC, Canada +1 819 821 8000, 1213

Department of Electrical and Computer Engineering, Université de Sherbrooke J1K 2R1, QC, Canada +1 819 821 8000, 2109

[email protected]

Soumaya.Cherkaoui@ USherbrooke.ca

communications. In this kind of networks, wireless devices can communicate with each other in the absence of a fixed infrastructure. Furthermore, ad hoc networks usually consist of a set of nodes that communicate over wireless links without the use of a central control, which creates a high level of flexibility to users.

ABSTRACT This paper explores the use of fuzzy logic for threshold buffers management in wireless ad hoc networks. This exploration is useful first, because of the dynamic nature of buffer occupancy and congestion at a node; second, because of the uncertainty of information in wireless ad hoc networks due to network mobility. The notion of threshold is practical for discarding data packets and adapting the traffic service depending on the occupancy of buffers. The threshold function has a significant influence on the performance of networks in terms of both packets average delay and throughput. We propose a fuzzy logic approach for threshold selection named (FuzzyCCG) in order to enhance the control of congestion. FuzzyCCG was studied under different mobility, channel, and traffic conditions. The results of simulations confirm that the proposed model can achieve low and stable end-to-end delay under different network scalability and mobility conditions. FuzzyCCG promises to be an efficient tool for reducing the delay of multimedia applications in wireless ad hoc networks.

However, ad hoc networks pose a great challenge to multimedia applications networking. Not only ad hoc networks inherit the classical problems of wireless and mobile communications such as bandwidth optimization and power control issues; the multihop nature of a hoc networks and the lack of a fixed infrastructure also introduce new research problems such as insuring some Quality of Service (QoS) transmission over highly dynamic topologies. The notion QoS is of central of importance to support any multimedia application. In ad hoc networks, this issue is particularly challenging because there is no fixed infrastructure and the topology is frequently changing due to node mobility. Consequently, links are constantly established and broken. The availability and quality of a link further fluctuates due to channel fading and interference from other transmitting devices. Various approaches and protocols have been proposed to address QoS ad hoc networking problem. Multiple efforts are in fact under way within academic and industrial research projects and the Internet Engineering Task Force. In section II, we give a brief description of the existing works.

Categories and Subject Descriptors C.2.1 [Computer Communication Networks]: Network Architecture and Design – wireless communication, network communication.

General Terms Algorithms, Management, Performance, Design.

In this paper, we explore a new QoS approach for wireless ad hoc networks. The proposed approach named FuzzyCCG is a fuzzy logic technique for improving the control of congestion. FuzzyCCG performs buffer thresholds management in wireless ad hoc networks. The notion of threshold is practical for discarding some data packets and adapting the traffic service to the occupancy of buffers. The threshold function has a significant influence on the performance of a network in terms of both packets average delay and throughput. Therefore, the selection of a particular threshold may be decisive for an adequate congestion control. FuzzyCCG exploration is useful first, because of the dynamic nature of buffer occupancy and congestion at a node; second, because of the uncertain nature of information in wireless ad hoc networks due to network mobility. In order to deal with the dynamic buffer occupancy and the uncertain and imprecise nature of information about the network in ad hoc networks information, we propose the use of a fuzzy logic approach, FuzzyCCG, for

Keywords Fuzzy logic, QoS, wireless mobile ad hoc networks, congestion control, threshold management.

1. INTRODUCTION Ad hoc networks are a new paradigm in the evolution of wireless

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Q2SWinet’05, October 13, 2005, Montreal, Quebec, Canada. Copyright 2005 ACM 1-59593-241-0/05/0010...$5.00.

105

It is important to note that the ability to provide QoS depends also on how well the resources are managed at the MAC layer. Some of the works cited above used generic QoS measures and are not tuned to a particular MAC layer [2], [11]. Some others use CDMA to eliminate the interference between different transmissions [8], [16]. The authors in [17] have introduced an on-demand, link-state, multi-path QoS routing protocol which collects information of link bandwidth from source to destination under the CDMA-over-TDMA channel model. Similarly, CDMAover-TDMA channel model has been adopted in [16] by using the notion of a time slot on a link to calculate the end-to-end path bandwidth. The same model has been used for calculating the end-to-end path bandwidth to develop on-demand QoS routing [8] and DSVD based QoS routing [16].

threshold selection. We study FuzzyCCG performances under different network conditions in terms of mobility and scalability. The results of simulations shown in Section IV confirm that FuzzyCCG promises to be an efficient tool for reducing the delay of multimedia applications in wireless ad hoc networks. This paper is organized as follows: in Section II, we discuss related works. Section III describes FuzzyCCG fuzzy logic approach for threshold management. Section IV shows the simulation results of the FuzzyCCG approach under different network conditions and traffic loads. Finally, Section V concludes the paper.

2. RELATED WORKS Research in the area of ad hoc networks is proceeding in both academia and industry under military and commercial sponsorship. Recently, there have been considerable efforts in the area of supporting QoS in mobile ad hoc networks (MANETs). The works that exist tend to be based on distributed scheduling algorithms that address QoS routing issues, QoS-based medium access controllers, rescheduling when the network topology changes, and fairness issues. The works in [7]-[15] have studied the QoS routing issue. In [7], we have proposed a flexible QoS routing protocol (AQOPC) based on multi-service classes and multi-path schemes. It provides information about the state of bandwidth, end-to-end delay and hop count in the network. AQOPC performs an accurate admission control and a good use of network resources by calculating multiple paths and generating the needed service classes to support different QoS user requirements. In [13], a core-extraction distributed ad hoc routing (CEDAR) algorithm is proposed that uses core extraction, link state propagation, and route computation to support QOS in wireless ad hoc networks. In [8], the authors have addressed the problem of supporting real-time communications in a multihop mobile network using QoS routing that permits bandwidth calculation and slot reservation. This protocol can be applied to two main scenarios: multimedia ad hoc wireless networks and multihop extensions of wireless ATM networks. The ad hoc QoS on-demand routing (AQOR) is discussed in [9], which integrates signaling functions for resource reservation and QoS maintenance at per-flow granularity. Some works such those described in [8] and [13] have proposed table-driven routing approaches for QoS support. However, their performances are low compared to reactive approaches because of the problem of stale route information [14]. A link-state QoS routing protocol for ad hoc networks (QOLSR) was proposed in [10] with the aim of implementing QoS functionality while dealing with limited available resources in a dynamic environment. A ticket-based QoS routing protocol was proposed in [11]. This protocol is based on a model which assumes that the bandwidth of a link can be determined independently of its neighboring links. Using the same model, [12] proposes a QoS multi-path routing protocol based on a ticket-distribution scheme to satisfy bandwidth constraints. Unfortunately, these schemes do not consider radio interference problems. The above-discussed QoS routing protocols can also be classified into two schemes: source routing and distributed routing. Most of the existing distributed algorithms (e.g., [15]) require the maintaining of a global network state at every node, which may cause the scalability problem. On the other hand, the source routing schemes such as [14] suffer from problems of scalability and frequent updates of the state of the network.

SWAN [1], FQMM [6], and INSINGIA [2] are the most noteworthy QoS models attempting to establish comprehensive QoS solutions for MANETs. SWAN proposes a service differentiation in stateless wireless ad hoc networks by using distributed control algorithms. It relies on feedback from the MAC layer as a measure of congestion in the network by using a mechanism of rate control and source-based admission control. It promotes a rate control system that can be used at each node to treat traffic either as real-time or best-effort traffic. However, one of the drawbacks of SWAN is that it does not offer a feasible mechanism for calculating the threshold rate to limit any excessive delay that might be experienced [18]. SWAN also uses merely two levels of services: real-time and best-effort traffic. INSIGNIA, such as SWAN, is an intranet QoS model providing services that have to be mapped to either per-flow or per-class services for wireless ad hoc networks. The main goal of INSIGNIA is to provide adaptive QoS guarantees for real-time traffic. It employs an in-band signaling system that supports fast reservation, restoration, and adaptation algorithms. Three levels of services are implemented: best-effort, minimum, and maximum. The bandwidth is however the only QoS parameter used in INSIGNIA. Another ad hoc QoS model, FQMM, is a hybrid approach combining the advantages of per-class granularity of DiffServ with the per-flow granularity of IntServ. It tries to preserve the per-flow granularity for a small portion of traffic in MANETs, given that a large amount of the traffic belongs to per aggregate of flows, that is, per-class granularity. FQMM offers a good solution for small- and medium-size ad hoc network, but it is not suitable for large networks. In [5], we have proposed an intelligent QoS model, named GQOS, \with service differentiation based on neural networks in mobile ad hoc networks. The main objective was to satisfy some QoS requirements, especially the reduction of end-to-end delay, in networks whose topologies change at low to medium rate. GQOS is composed of a kernel plan which ensures basic functions of routing and QoS support control, and an intelligent learning plan which ensures the training of GQOS kernel operations by using a multilayered feedforward neural network (MFNN). The advantage of using a neural network algorithm is the learning of different operations performed by the kernel and the subsequent reduction of the processing time in the network. However, the learning process is CPU consuming. In [4], we have explored the use of a fuzzy logic semi-stateless QoS approach for service differentiation in wireless ad hoc networks. The proposed model named

106

On the other hand, most of events occurring in an ad hoc network are dynamic and random; therefore manually predefining a value for threshold is not suitable. In addition, it is important to note that the rate of packets arriving on a particular node is not static. The threshold value divides the buffer into an “admitted” part and a “no-admitted” part. Let consider that the threshold of the buffer shown in Figure 1 is equal to 60%. In this scheme, the occupancy level may range from 0 to 60%. When the buffer occupancy is superior to 60%, no incoming packets are accepted in the buffer. Therefore, the change in decision making from “admit state” to “no-admit state” is performed from 60-61%. This means that a small variation in the buffer occupancy may influence the decision making of incoming packets.

FuzzyMARS includes a set of mechanisms: admission control for real-time traffic, a fuzzy logic system for best-effort traffic regulation, and three schemes for real-time traffic regulation. FuzzyMARS architecture support both real-time UDP traffic and best-effort UDP and TCP traffic. The resulted simulations have shown the benefits of using the proposed fuzzy logic semistateless model. The average delay obtained is quite stable and low under different channel conditions, traffic scalability, and mobility scenarios. However, in FuzzyMARS some bandwidth loss was experienced in overall network capacity. This may be due to the fact that no specific buffer management was used in FuzzyMARS. The model proposed here can be considered as an augmented FuzzyMARS, where we use fuzzy logic for buffer threshold management.

Full

3. FUZZY LOGIC APPROACH FOR THRESHOLD MANAGEMENT

Empty 60%

Figure 1. Classical buffer scheme.

Fuzzy logic theory [23]-[25] was first introduced as a tool for modeling the uncertainly of natural language, and has been commonly employed for supporting intelligent systems. This technology has proven efficiency in various applications such as decision support and intelligent control, especially where a system is difficult to be characterized. A fuzzy logic system considers basically three steps: fuzzification, rules evaluation, and deffuzification. The first step is responsible for mapping discrete (called also crisp) input data into proper values in the fuzzy logic space. For that end, membership functions (fuzzy sets) are used to provide smooth transitions from false to true (0 to 1). The second step performs reasoning on the input data by following predefined fuzzy rules. Once the input data are processed by fuzzy reasoning, the deffuzification takes the task of converting back these input data into crisp values.

0.9 0.7 Admitted packets Uadm

0.5 0.3 0.1

10%

20%

30%

40%

50%

60% 70%

80%

90% 100%

Figure 2. Fuzzy Buffer occupancy scheme.

In FuzzyCCG, we attempt to extend the two-discrete states “admit” and “no-admit” of the buffer occupancy by using fuzzy logic. The aim of introducing fuzzy logic is to develop a more realistic representation of buffer occupancy that helps to offer an efficient decision making. Hence, the definition of “buffer occupancy” will consider the two fuzzy cases of “getting full” and “not getting full”, rather than “admit” and “no-admit” in the existing approaches. This fuzzy representation replaces the twodiscrete sets by a continuous set membership, and performs small gradual transitions between different states of buffer occupancy.

Given the results found with FuzzyMARS, fuzzy logic promises to also offer an efficient tool for buffer management by using adequate thresholds that deal with wireless ad hoc network dynamics. Also, fuzzy logic has been successfully applied to the queue management in cell-switching networks [22]. Nevertheless, to the best of our knowledge, this is the first work that uses fuzzy logic for buffer management in MANETs. We aim to apply a fuzzy technique based on fuzzy sets theory. The later extends the classical logic set {0, 1} to use linguistic variables (e.g. full buffer, merely full buffer, empty buffer). Using fuzzy logic, we investigate the fuzzy thresholds ability to adapt to the dynamic conditions over the classical inflexible thresholds.

The fuzzy membership function aims to determine the fuzzy threshold depending on the fullness of the buffer. Several membership functions may be used for that purpose: “triangular”, trapezoidal”, or “sigmoid” function. These functions can give a representation about the buffer fullness level. In FuzzyCCG, we used the sigmoid membership function. This choice is based on the fact that this function would reflect well the dynamic occupancy of the buffers that we want to model.

It is observed that the classical thresholds are excessively restrictive, because the selection of threshold is based on a single value. Thus, the utilization of a buffer may be either “poor” or “surcharged”. When the selected value is small (e.g. 30% of capacity), then the admission of new packets is possible only when the buffer occupancy is low. This means a poor utilization of the buffer; since most of incoming packets are rejected even if the buffer is almost unfilled. On the other side, when the selected value is big (e.g. 90% of capacity), problems may happen when the bursty traffic is used. The transmission of packets generated by a bursty traffic is very changing. It can vary from small to “near-peak” rate in a short period of time.

As shown in Figure 2, the admit membership function is inversely proportional to the occupancy fullness level of buffer. Thus, when the occupancy fullness is small, the value of the admit membership function is big. At higher fullness occupancy levels, the admit membership function value becomes small. When the value of the “no-admit” membership function is getting big, then only a small quantity of packets will be permitted to enter the buffer. In Figure 2, the value of the membership function is

107

represented by the symbol. as follows:

u adm

msec. In contrast, the average delay in the original model grows from 7 to 31 msec as the number of TCP flows increases from 2 to 12 flows. Hence, the gain achieved by FuzzyCCG in terms of the average end-to-end delay, is by about 74-92%. In Figure 5, we observe that the average delay in FuzzyMARS grows slowly with the increasing number of TCP flows, and it remains almost less than 3 msec. We observe that the average delay of the TCP traffic in FuzzyCCG is almost the same as in FuzzyMARS (FuzzyMARS outperforms FuzzyCCG by about 3%). Figure 7 shows the average end-to-end delay in both FuzzyCCG and SWAN models. It is shown that the average delay is almost inferior to 3 msec in the proposed model, whereas in SWAN model the average delay is around 5 msec. This means that we can achieve a gain of about 49% in terms of average delay in FuzzyCCG.

The fuzzy rules associated are

<< When the value of the admit membership function is big, then increase the accepted incoming packets into buffer >> << When the value of the admit membership function is small, then reduce the accepted incoming packets into buffer >> The previous fuzzy rules are illustrated by Figure 2. The rejection of packets is controlled based on the degree of fullness of the buffer. For instance, when the buffer is occupied at 40%, this means that the value of u adm is about 0.7 (i.e. the amount of packets admitted is about 70%). Then, about “30%” of incoming packets will be not admitted. Note that the fuzzy threshold approach covers the continuous set of values representing possible

We observe in Figure 4, 6 and 8 the impact of growing number of TCP flows on the average throughput over the different models of simulation. The average throughput of the TCP traffic in FuzzyCCG is almost the same as in the original model, as shown in Figure 4. At a lower number of TCP flows, the average throughput in the original is superior to that in FuzzyCCG. A similar result is observed in Figure 6 between FuzzyCCG and SWAN. On the other hand, Figure 8 shows that the FuzzyCCG outperforms FuzzyMARS in terms of average throughput of the TCP traffic by about 21% at cost of 3% decrease in the average delay.

buffer occupancy (i.e. from 0 to u adm ). This is opposite to the classical threshold approaches that hold only one predefined single value. Therefore, fuzzy logic adds more flexibility to the threshold selection.

4. PERFORMANCE EVAULATION We integrated the fuzzy buffer threshold management with FuzzyMARS. The aim is to evaluate whether the results will outperform other existing models such as SWAN, and also asses if the integrated buffer management adds more performances to FuzzyMARS. The performance evaluation of the proposed QoS model is studied with the scalable ns-2 simulator. Each mobile host has a transmission range of 250 meters and shares an 11 Mbps radio channel with its neighboring nodes. We compare the performance of FuzzyCCG with the ‘original model’, FuzzyMARS as described in our previous work [4] and the SWAN model described in [1]. We use the word ‘original model’ to refer to IEEE 802.11 wireless networks without FuzzyCCG mechanisms.

0,03

FuzzyCCG Or iginal

0,025 0,02 0,015 0,01 0,005 0 2

4

6

8

10

12

N u mb e r o f TC P f l ows

In order to better understand the properties of the FuzzyCCG regulation, the simulation considers multiple scenarios of realtime and TCP best-effort traffic. The real-time traffic is modeled as 4 voice and 4 video flows. The TCP traffic is modeled as a mixture of FTP and Web traffic. Web traffic represents microflows, whereas FTP traffic corresponds to macro-flows. The video and voice flows representing real-time traffic are active and monitored for the duration of 100 seconds. Video traffic is modeled as 200 Kbps constant rate traffic with a packet size of 512 bytes. Voice traffic is modeled as 32 Kbps constant rate traffic with a packet size of 80 bytes. The simulation considers a multihop network of 50 mobile nodes. The network area has a rectangular shape of 1500m x 300m that minimizes the effect of network partitioning. The AODV protocol [19] is chosen as the routing protocol.

Figure 3. Average delay in the original and FuzzyCCG models vs. number of TCP flows.

FuzzyCCG

1200

Or iginal 900

600

300

0 2

Figures 3-8 present the scalability impact of the increasing number of TCP flows on the average end-to-end delay and throughput of traffic. Figure 3 illustrates a significant difference in terms of the average delay between FuzzyCCG and the original model. The average delay in FuzzyCCG grows slowly with the increasing number of TCP flows, and it remains between 2 and 3

4

6 8 Number of TCP f l ows

10

12

Figure 4. Average throughput in the original and FuzzyCCG models vs. number of TCP flows.

108

Therefore by adopting the FuzzyCCG mechanisms, we can achieve a reduction in the average end-to-end delay by about 7492%, 49%, and 21% in comparison respectively to the original model, SWAN, and FuzzyMARS, with almost the same average throughput. This allows FuzzyCCG to support efficiently multimedia applications.

FuzzyCCG FuzzyMARS

0,006

0,004

0,002

The impact of mobility on the performances of FuzzyCCG is investigated in Figures 9-14. The real-time traffic is modeled in the same manner as discussed previously. The best-effort TCP flows consists of 5 web flows and 5 FTP flows. The random waypoint mobility model [20] is implemented at each node in the network. In the beginning, the nodes are randomly placed in the area. Then, each mobile node selects a random destination and moves with a random speed up to a maximum speed of 20m/s.

0 2

4

6 8 Number of TCP f lows

10

12

Figure 5. Average delay in FuzzyCCG and FuzzyMARS models vs. number of TCP flows.

900 FuzzyCCG

800

FuzzyMARS

700 600 500 400

0,04

300

FuzzyCCG Or iginal

200

0,032

100

0,024

0 2

4

6

8

10

12

0,016

N umbe r of T C P f l ows

0,008

Figure 6. Average throughput in FuzzyCCG and

0

FuzzyMARS models vs. number of TCP flows.

300

200

120

60

30

M obi l i t y ( P a use t i m e , se c )

Figure 9. Average delay in the original 0,02

and FuzzyCCG models vs. mobility.

FuzzyCCG SWAN

0,015

640

0,01

560 480

0,005

400 320

0 2

4

6

8

10

240

12

N um be r of TC P f l ows

FuzzyCCG

160

Or i gi nal

80

Figure 7. Average delay in FuzzyCCG and SWAN models vs. number of TCP flows.

0 300

200

120

60

30

Mobilit y ( Pause t ime, sec)

Figure 10. Average throughput in the original and FuzzyCCG models vs. mobility. 1200

FuzzyCCG

1000

SWAN

800 600 400 200 0 2

4

6

8

10

12

N umbe r of T C P f l ows

Figure 8. Average throughput in FuzzyCCG and SWAN models vs. number of TCP flows.

109

0,01

FuzzyCCG

640 0,008

FuzzyMARS

560 480

0,006

400 320

0,004 0,002

FuzzyCCG

240

SWAN

160 80

0 300

200

120

60

0

30

300

200

120

60

30

M obi l i t y ( P a us e t i me , s e c )

M obi l i t y ( pa us e t i me , s e c )

Figure 11 Average delay in FuzzyCCG and

Figure 14. Average throughput in FuzzyCCG

SWAN models vs. mobility.

and FuzzyMARS models vs. mobility.

After reaching the destination, the node will stay there for a given “pause time” then starts to move towards another destination. This process is repeated during for the duration of simulation.

Figure 11 illustrates that the average end-to-end delay in FuzzyMARS increases slowly and it grows only for the highest mobility scenarios. The average delay offered by FuzzyMARS is about 2% better than that FuzzyCCG. However, it is shown in Figure 12 that for different mobility scenarios, the throughput in FuzzyCCG is better than in FuzzyMARS model. FuzzyCCG acts better in terms of throughput by about 49% than the FuzzyMARS.

Figure 9 shows that the average end-to-end delay in FuzzyCCG increases slowly. The average delay in the proposed model remains almost less than 5.4 msec, whereas the average delay in the original model grows from 25 to 38 msec. This means that the proposed FuzzyCCG achieves a reduction in terms of average delay by about 79-87%. On the other hand, it is observed in the Figure 10 that the throughput of TCP best-effort traffic decreases slowly in the original model as the mobility increases. The average throughput in FuzzyCCG is superior to that of the original model by about 33% for different mobility scenarios.

We observe in Figure 13, the average end-to-end delay with different mobility scenarios in both FuzzyCCG and SWAN models. It is observed that the average delay of traffic in FuzzyCCG increases slowly as the mobility increases. For different mobility scenarios, the average delay offered by FuzzyCCG is about 10-36% better than that offered by SWAN. It is shown in Figure 14 that for different mobility scenarios, the throughput in FuzzyCCG is better than in SWAN model by about 43%. During the mobility of nodes, some flows are dropped in both SWAN and FuzzyCCG models because of the difficulty in capturing the dynamics of the environment in the ad hoc network.

FuzzyCCG

700

SWAN

600 500 400 300

5. CONCLUSION

200 100

In this paper, we proposed the FuzzyCCG approach which is a new QoS approach based on fuzzy logic for mobile ad hoc networks. FuzzyCCG explores how fuzzy logic can enhance the buffer threshold management in wireless ad hoc networks. FuzzyCCG exploration is useful because of the importance of the threshold notion for discarding of data packets when necessary and adapting the traffic service to the occupancy of buffers. In addition, the threshold function has a significant influence on the performance of the network in terms of both packets average delay and throughput. Therefore, the selection of a particular threshold may be decisive to QoS management. The implemented technique proved to be efficient and scalable. The performances evaluation of FuzzyCCG was studied using the ns-2 simulator, under diverse mobility and traffic conditions. In terms of traffic scalability, the simulation has shown that we can achieve a reduction in terms of average end-to-end delay by about 74-92% and 49% in comparison to respectively, the original (i.e. IEEE 802.11 wireless networks) and SWAN models, with almost the same throughput. Furthermore, FuzzyCCG outperforms FuzzyMARS in terms of average throughput by about 21% at a cost of 3% decrease in the average delay. On the other side, the proposed model proves better performances over the original, FuzzyMARS, and SWAN models under various mobility

0 300

200

120

60

30

Mobilit y (Pause t ime, sec)

Figure 12. Average throughput in FuzzyCCG and SWAN models vs. mobility.

0,006 0,005 0,004 0,003 FuzzyCCG

0,002

FuzzyMARS 0,001 0 300

200

120

60

30

M obi l i t y ( pa us e t i me , s e c )

Figure 13. Average delay in FuzzyCCG and FuzzyMARS models vs. mobility.

110

[11] S. Chen and K. Nahrstedt. “Distributed Quality-of-Service in Ad Hoc Networks”, IEEE Journal on Selected Areas in Communications, vol.17, no. 8, Aug., 1999.

scenarios. The performance results of FuzzyCCG confirm that the presented fuzzy logic approach promises to support efficiently the multimedia applications in wireless ad hoc networks by reducing the traffic delay while keeping a high throughput.

6. ACKNOWLEDGMENTS

[12] W.-H. Liao, Y.-C. Tseng, J.-P. Sheu, and S.-L. Wang. “A Multi-Path QoS Routing Protocol in a Wireless Mobile Ad Hoc Network”, In the proceedings of IEEE ICN’01, Int. Conference on Networking, Part II, pp. 158–167, July, 2001.

This research is supported financially by the Natural Sciences and Engineering Research Council of Canada (NSERC), and the Canadian Foundation for Innovation (CFI).

[13] R. Sivakumar, et al., “CEDAR: a core extraction distributed ad hoc routing algorithm”, IEEE Journal on Selected Areas in Communications, vol. 17, no. 8, pp. 1454-1465, 1999.

7. .REFERENCES

[14] S. Chen and K. Nahrstedt, “On finding multi-constrained paths”, IEEE Int. Conference on Communication, 874 -879 June 98.

[1] G.H. Ahn, A. T. Campbell, A. Veres, and L. H. Sun, “SWAN: Service Differentiation in Stateless Wireless Ad Hoc Networks”, IEEE INFOCOM 2002.

[15] Z. Wang and J. Crowcroft, “QoS routing for supporting resource reservation”, IEEE Journal on Selected Areas in Communications, vol. 14, no. 7, 1996.

[2] S.-B. Lee, G.-S. Ahn, X. Zhang, and A.T. Campbell, “INSIGNIA: An IP-Based Quality of Service Framework for Mobile Ad Hoc Networks,” Journal of Parallel and Distributed Computing (Academic Press), special issue on wireless and mobile computing and communications, vol. 60, no. 4, pp. 374-406, Apr. 2000.

[16] C.-R. Lin. “On-Demand QoS Routing in Multihop Mobile Networks”, In Proc. of IEEE INFOCOM 2001, pp. 1735– 1744, April 2001. [17] Y. Chen, Y. Tseng, J. Sheu, and P. Kuo, “On-Demand, LinkState, Multi-Path QoS Routing in a Wireless Mobile Ad-Hoc Network”, European wireless, Florence, Feb. 2002.

[3] J.L. Sobrinho and A.S. Krishnakumar, “Quality-of-Service in Ad Hoc Carrier Sense Multiple Access Networks,” IEEE Journal on Selected Areas in Communication, vol. 17, no. 8, pp. 1353-1368, Aug. 1999.

[18] Y. L. Morgan, T. Kunz, “PYLON: An architectural framework for ad-hoc QoS interconnectivity with access domains”, HICSS’03 pres, Hawaii, USA, Jan. 2003. [19] C.E. Perkins, E.M. Royer, “Ad-hoc on-demand distance vector routing”, In the Proc. of the Second IEEE Workshop on Mobile Computing Systems and Applications, pp.90 – 100. New Orleans, LA, Feb. 1999.

[4] L. Khoukhi, S. Cherkaoui, “FuzzyMARS: A Fuzzy Logic Approach with Service Differentiation for Wireless Ad hoc Networks”, IEEE International Conference on Wireless Networks, Communications, and Mobile Computing WirelessCom2005, June 13-16, 2005.

[20] J. Broch, D. Maltz, D. Johnson, Y.-C. Hu, and J. Jetcheva, “A Performance Comparison of Multihop Wireless Ad Hoc Network Routing Protocols,” In the proceedings of ACM/IEEE Int’l Conf. Mobile Computing and Networking (MobiCom), Oct. 1998.

[5] L. Khoukhi, S. Cherkaoui, “A Quality of Service Approach Based on Neural Networks for Mobile Ad hoc Networks”, IEEE-IFIP International Conference on Wireless and Optical Communications Networks WOCN 2005, Dubai, UAE, March 6 - 8, 2005.

[21] F. Kevin, V. Kannan, “NS notes and documentation,” in The VINT Project, UC Berkely, LBL, USC/ISI, and Xerox PARC, 1997.

[6] H. Xiao, W. K.G. Seah, A. Lo, and K. Chaing, “Flexible QoS Model for Mobile Ad-hoc Networks”, IEEE Vehicular Technology Conference, Vol. 1, pp 445-449, Tokyo, May 2000.

[22] A. R. Bonde and S. Ghosh, “A comparative study of fuzzy versus Fixed thresholds for robust queue management in cell-switching networks,” IEEE Trans. Networking, vol. 2, pp. 337–344, Aug. 1994.

[7] L. Khoukhi, S. Cherkaoui, “Flexible QoS Routing Protocol for Mobile Ad Hoc Networks”, In Proc. of the 11th IEEE International Conference on Telecommunication (ICT2004), Brazil, Aug. 2004.

[23] X. Wang, “A Course in Fuzzy Systems and Control”, Upper Saddle River, NJ: Prentice-Hall, 1997.

[8] C. R. Lin and J.-S. Liu. “QoS Routing in Ad Hoc Wireless Networks”, IEEE Journal on Selected Areas in Communication, Vol. 17, No. 8, 1426–1438, 1999.

[24] C.C Lee, “Fuzzy logic in control systems: fuzzy logic controller- part I and II,” IEEE transactions on Systems, Man, and Cybernetics, vol. 20, no. 2, 1990.

[9] Qi Xue and Aura Ganz. “Ad hoc QoS on-demand routing (AQOR) in mobile ad hoc networks”, Journal of Parallel and Distributed Computing, Elsevier Science, USA, 2003.

[25] L. A. Zadeh, “fuzzy logic = computing with words”, IEEE Transactions on fuzzy systems, vol. 4, no2, pp. 104-111, 1996.

[10] A. Munaretto, H. Badis, K. Al Agha and G. Pujolle, “A Link-state QoS Routing Protocol for Ad Hoc Networks”, In the proceedings of IEEE MWCN02, Stockholm, Sept.2002.

111