Congestion detection for video traffic


Published on

1 Comment
  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Congestion detection for video traffic

  1. 1. Congestion Detection for Video Traffic in Wireless Sensor Networks Hemmat Sheikhi Mousa Dashti Mehdi DehghanDept. of Computer Eng., Amirkabir Dept. of Computer Eng., Amirkabir Dept. of Computer Eng., Amirkabir University of Technology University of Technology University of Technology Tehran, Iran Tehran, Iran Tehran, Iran— Congestion control mechanisms include three phases: retransmitted packet burdening excessive delay may becongestion detection, congestion notification and rate adjustment. unacceptable. So it is obvious in WMSNs that achievingSo far diverse congestion detection methods for sensor networks reliability by congestion control approach is of great priority.are proposed. In this paper we introduce numerous congestiondetection parameters and examine them in various respects; Among multimedia applications those that have videofinally we choose one of them as the best parameter for video traffic due to high amount of data rate and also because oftraffic in wireless sensor networks. Some of intended criteria for burst nature are more in congestion danger (Typicallycomparing the parameters are cost, relation to quality of video, transmission rate in sensor networks is 40 kbps but audiolocality or being global in the network, accuracy and speed of traffic rate in constant data rate is 64 kbps and burst of videocongestion detection. We simulated and concluded that average traffic is about 500 kbps [4]). So any proposed mechanism fordelay is the most suitable parameter for congestion detection in congestion control in WMSNs must be respondent to videothese networks. applications requirements and complying this condition will be suitable to other multimedia applications. Keywords: Wireless Networks; Sensor; Congestion detection;Video; Delay. Congestion control is conducted in three phases: congestion detection, congestion notification, and rate adjustment. In this paper we focus on first phase of congestion I. INTRODUCTION control that is congestion detection. In section II we introduce Advent of new technologies in sensors, camera and main congestion detection parameters and then we comparemicrophones in smaller scales and with lower energy them in various respects.consumption led to the design of wireless sensors with theability to sense their around environment. Nowadays there is II. CONGESTION DETECTIONwidespread research in this area and new applications of thesesensors are becoming popular. An example of these So far, various methods are proposed in WSNs each ofapplications is utilization of these sensors to monitor around which using one parameter for congestion detection. Selectingenvironment that led to the advent of Wireless Multimedia this parameter is based on various factors, some of whichSensor Networks (WMSNs) [1]. Designing these networks has being: network structure, data transmission rate, trafficmany challenges such as nature of wireless media and pattern, congestion probability, type of network applicationsmultimedia information transmission. Consequently traditional and QoS requirements of them, congestion effect onmechanisms for network layers are no longer acceptable or applications and network resources [3].applicable for these networks. Transport layer is one of the Table 1 shows congestion detection parameters along withmain layers in WMSNs which has greatly influenced the their related protocols. Besides, it is illustrated that whichoverall performance of received packets. This is because of parameter is applied to what type of nodes. In each method tolimited bandwidth, high data transmission rate, burst nature of make out which node is in charge of congestion detection isthis multimedia traffic and high effect of congestion on these dependent on the parameter nature. For example using queuepackets [2]. Proposed methods for transport layer of sensor length is tapped only in intermediate nodes and in networksnetworks are classified in two categories [3]: with end to end retransmission, retransmission time is only used in Sink to detect congestion. Naturally parameters that • Methods that provide reliability by retransmission are applicable to all of nodes are more flexible since approach depending on network conditions and requirements it is • Methods that reliability is achieved by congestion possible to determine location of detecting congestion. For control example, if speed of congestion detection is at question, we Retransmission causes excess power consumption and is can do it in intermediate nodes and if reducing load of sensornot convenient for WMSNs. On the other hand because nodes is aimed, it is possible to use sink node for congestionmultimedia applications are sensitive to delay, reception of detection. This work is partly sponsored by a research grant awarded by Iran Telecommunication Research Center (ITRC). 978-1-61284-459-6/11/$26.00 ©2011 IEEE 1127
  2. 2. TABLE I. CONGESTION DETECTION PARAMETERS first classes consume large amount of energy and so are not Parameter Location Protocol suitable for WMSNs. STCP[5], Fusion[6], Internal queue length Intermediate nodes Siphon[7], DECbit[8], Some of congestion detection methods require ESRT[9] synchronization. In this paper we have not considered Inter packet arrival Sink and intermediate synchronization between sensor nodes. The methods described PCCP[10] time nodes here may be used without synchronization between nodes byService time of packets Intermediate nodes PCCP, CCF[11] some heuristics. In what follows we examine remaining Load existing in parameters in two last classed of the table. Intermediate nodes CODA[12] channel Sink and SourceRate of logging packets ESRT, CODA B. Effect of congestion detection on quality of received video nodes Sensing packets( One of the main criteria that are important for selecting without congestion congestion detection parameters is the way that parameter Intermediate nodes ARC[13] detection or notification) affects the quality of received video in Sink. The more that parameter affects the quality of video; the better is to use it inRetransmission time of packets Sink RCRT[14] WMSNs. As we know delay, jitter and packet loss are metrics of quality of service. Among the parameters remaining from Sink and intermediate Delay nodes no protocol previous section, delay and jitter comply with quality of Sink and intermediate service. Other parameters such as queue length, service time Jitter no protocol of packets or inter-arrival time of packets are indirectly affect nodes Power variance Intermediate nodes no protocol the quality of service. Reducing these parameters causes the decrease of delay and as a result enhancement of the quality of received video. Every protocol that is proposed for WMSNs In what follows, all of congestion detection parameters are should take these parameters into account. By bringing theseexamined and eventually we select the most suitable parameters into consideration we can provide quality ofparameter. Metrics of comparison is the cost, impact on service of applications in transport layer. For example if delayquality of video, locality or being global in the network and is used for congestion detection, threshold of delay can befalse positivity and speed of congestion detection. adjusted to comply to play-out time in receiver. In TABLE III. each of the remaining parameters of previous section is classified based on effect on quality of service.A. Cost of congestion detection One of the comparison metrics in congestion detection is C. Locality or globality of parameterthe cost. Some methods have overhead cost. Method with Among parameters of congestion detection, some oflower cost is most convenient in sensor networks. This cost is parameters detect congestion only through local informationevaluated in two aspects: power overhead and processing that is available in node. For example when queue length isoverhead. used, each of the nodes detects congestion only based on its Processing overhead: According to TABLE I. the own queue length. But some other parameters use moreparameters that are involved in intermediate nodes such as information and do not rely on their own information. Amongqueue length, channel load or power variance have more them, delay parameters use the sum of delay of all nodes in theprocessing overhead because of large amount of load on path. Needless to say that parameters that use globalintermediate nodes. information are better than those using local information. In TABLE IV. we classified parameters based on this criteria. Power overhead: This cost is the amount of energy that isconsumed for congestion detection. These parameters are D. Congestion Misdetectionclassified in three categories: Another criterion for congestion detection parameter • Sensing the channel: Among methods that are listed assessment is that how accurately that parameter detects in the TABLE I. those that have the cost of sensing congestion. The more that parameter accurately detect channel have higher energy consumption and so they are not suitable for WMSNs. TABLE II. COST OF CONGESTION DETECTION • Using extra packets: Using retransmission time of dropped packets includes not only retransmission Congestion detection Cost request but also transmission of dropped packet. parameters These methods waste a great amount of energy for Sensing channel Exponential power overhead Retransmission time Extra packet transmit congestion detection in sensor nodes. Energy variance, queue length, • Low cost: Some methods do not necessitate extra service time of packets, overall Processing overhead (Intermediate cost for congestion detection. These methods are the service time, delay, delay nodes are also in charge of most suitable for congestion detection in WMSNs. variance, Inter arrival time of congestion detection) In TABLE II. all methods are classified based on cost. packetCongestion detection parameters that are classified under two Delay, jitter, inter arrival time of Low cost (only destination node is in packets charge of congestion detection) 1128
  3. 3. TABLE III. CONGESTION DETECTION PARAMETER AND QUALITY OF Davg = * Davg + (1- ) * d (1) VIDEO Javg = * Javg + (1- ) * (d – Davg) (2) Parameters Effect on quality of video Energy variance No effect In these formulas d is delay of current packet and is queue length, inter arrival time, Indirect effect weight that is assigned to delay in weighted average delay. service time delay, jitter Indirect effect Davg is average delay of packets of a flow. In above equation for computing average jitter instead of using absolute value of TABLE IV. LOCAL OR GLOBAL jitter, the unchanged jitter is used. This is because when congestion is terminated and delay is reduced using absolute Parameter Information Type jitter causes that average jitter is increased and a misdetectionqueue length, energy variance, service local of congestion is produced. If we do not use absolute value time delay, jitter, Inter arrival time global when congestion is passed average jitter is decreased. In the above equations the more be closed to 1 meanscongestion the more is convenient for this task. Misdetection that we gave a more weight to previous average delay. Sooccurs in two cases: upcoming congestion is not detected and average delay gained a slower pace than change of delay anda notified congestion is not an actual congestion that is going so reaction to congestion is slow. Advantage of this behaviorto occur. For investigating this issue firstly we should get is that when delay of some packets is not because oftraffic pattern and then we should consider change in congestion situation and is transient, this method does notcongestion detection parameters so that accuracy or have congestion misdetection. The more is close to 0, itmisdetection of them is recognized. means that more weight is used for delay of current packets. As we know video files with MPEG format consist of 3 So a small increase in delay of current packets increases thetypes of frames named B, P, I. Distance between two I frames average. In this case congestion is detected more rapidly but inis called GOP. B and P frames are between I frames. Number some cases there is a misdetection of a nonexistent congestion.of these frames depends on encoding of frames. Length of We know that by the advent of congestion in network,each type of frame and number of them in network are queue length in intermediate nodes is increased and as a resultdifferent. I frame has the largest length and when transmitting delay of packets is increased. Having a scrutiny in abovethis type of frame, rate of transmission is high. B and P frames equations we come to the conclusion that delay of each packethave lower length although B frame has lower length than P. has a direct effect on average delay. But difference of delay of Occurrence of congestion in network is proportionate to each packet versus average delay makes jitter. So by increasenumber of resources that at the same time or in a low interval of delay, average of delay grows quicker than average of jittertransmit I packets in one route. Variance of most of the and therefore congestion is detected and controlled quickercongestion detection parameters is proportionate to traffic consequently.pattern. This means that with increasing transmission rate, The other problem of average jitter is that when congestionvariance of them are increased and reducing rate decreases frequently occurs in network and network is often intheir variance. The only parameter that does not show this congestion state, average jitter is reduced instead of increasingbehavior is inter-arrival time of packets. This parameter is or remaining constant and as a result congestion is notuseful when inter transmission time of packets are equal and detected. On the other hand average delay in these situationsin such a condition it is recognized that if inter reception of always is above its threshold and always detects thethem changed we infer that there is a congestion in the path. congestion.But source node in video traffics transmit packet in differentintervals (packets belonging to different frame types) and so To sum up, average jitter is not a suitable parameter forsink cannot determine whether the interval between receiving video traffic congestion detection. In the following wepackets are due to congestion or for another reason. So this simulate a congested network to verify the discussion andparameter is not convenient for our video network. select the best congestion detection parameter in accuracy and quickness. The parameter that responds quicker to congestion Accordingly some parameters are more suitable to our is the most convenient. Remaining congestion detectionnetwork. These parameters are: delay, jitter, queue length and parameters are: average delay, average service time, queueservice time of packets. In the following we examine these length of nodes.parameters and then we select best of them. III. SIMULATIONE. Speed of congestion detection We use NS2[15] and Evalvid[16] tool in our simulation. Quick congestion control depends on two factors: quick Simulation parameters are shown in TABLE V. Five nodes arecongestion detection and suitable rate adjustment. One considered in our simulation arrangement of which areimportant criteria comparison among congestion detection depicted in Figure 1. Node 5 is sink. Initially consider thatmethods is that which method can detect congestion more node 1 sends packets of Foreman video file with MPEGinstantaneously. One of the most useful criteria is that which format. In Figure 2 we have shown change process of delayparameter has more change in case of network congestion. For parameter for node 5. Average service time and queue lengthexample comparing two parameters of delay and jitter we have for node 3 is also shown in Figures 3 and 4. We assume thatthe following averages for them. 1129
  4. 4. our network can tolerate one single burst and applicationswould not be affected in such a burst. But our network will notbe respondent if two or more flows simultaneously go to aburst. So packets will be late in sink or will be discarded. Thuscongestion will occur and we must detect it. We want to usethe parameter that detects it quicker and with moreprobability. For evaluating threshold value we use the followingmethod. Maximum amount of that parameter in case of onesingle burst of a flow will be our threshold. Now withincreasing simultaneous flows we investigate that whichparameter and when violates the threshold. In previous figures Figure 3. Average service time for 1 flowwe conceive that maximum amount of average delay for aflow is 64 milliseconds and maximum queue length for thesame number of flows is 6 and maximum average service timeis 16 milliseconds. We consider these values as our networkthreshold. TABLE V. SIMULATION PARAMETERS Simulation parameters Area 200mX200m Channel WirelessChannel Propagation Model TwoRayGround Energy Consumption Modelu EnergyModel Antenna OmniAntenna Bandwidth 5Mbps Figure 4. Queue length for 1 flow We start the simulation from scratch. This time both node 1 and 2 are sending simultaneously and because they use the same coding format they go burst together. Average delay of node 5, average service time and queue length for node 4 is depicted in figures 5, 6 and 7. In these figures we see the change in value of parameters with the increase of a simultaneous flows and occurrence of congestion. We observe that average delay only in a single congestion case does not violate its threshold. But queue length passes its threshold only 3 times and this threshold passing for service time is only 5 times. Both service time and queue length have similar change. But delay in occurrence of congestion have Figure 1. Network architecture further change and so better detects congestion in terms of both speed and accuracy. Figure 2. Average delay for 1 flow Figure 5. Average delay for 2 flows 1130
  5. 5. [3] C. Wang, K. Sohraby, "A Survey of Transport Protocols for Wireless Sensor Networks," IEEE Network 20 (2006) 34-40. [4] X. Zhu, B. Girod, "Distributed rate allocation for multistream video transmission over ad hoc networks," In Proc. of IEEE Intl. Conference on Image Processing (ICIP) 2005 Boston 157-160. [5] Y. G. Iyer, S. Gandham, S. Venkatesan, "STCP: A generic transport layer protocol for wireless sensor networks," In Proc. International Conference on Computer Communications and Networks (ICCCN) 2005 Houston 449-454. [6] B. Hull, K. Jamieson, H. Balakrishnan, "Mitigating congestion in wireless sensor networks ," In Proc. of the First International Conference on Embedded Networked Sensor Systems (Sensys) 2004 Baltimore 266-279. Figure 6. Average Service time for 2 flows [7] C. Wan, S. Eisenman, A. Campbell, j. Crowcroft, "Siphon: Overload traffic management using multi-radio virtual sinks in sensor networks", In Proc. of the 3rd international conference on Embedded networked sensor systems (SenSys) 2005 California 116-129. [8] K. Ramakrishnan, R. Jain, "A binary feedback scheme for congestion avoidance in computer networks," Computer Communication 25 (1995) 138-156. [9] O. Akan, I.F. Akyildiz, "Event-to-sink reliable transport in wireless sensor networks," IEEE/ACM Transactions on Networking 13 (2005) 1003–1017. [10] C. Wang, K. Sohraby, V. Lawrence, B. Li, Y. Hu, "Priority-based congestion control in wireless sensor networks," IEEE International Conference on Sensor Networks, Ubiquitous, and Trustworthy Computing (SUTC) 2006 Karlsruhe 22-29. [11] C.-T. Ee and R. Bajcsy, “Congestion control and fairness for many-to- one routing in sensor networks," In Proc. of the Second International Figure 7. Queue Length for 2 flows Conference on Embedded Networked Sensor Systems (Sensys) 2004 Baltimore 148-161. [12] C.Y. Wan, S. B. Eisenman, A. T. Campbell, "CODA: Congestion IV. CONCLUTION detection and avoidance in sensor networks," In Proc. of the First International Conference on Embedded Networked Sensor Systems According to what preceded in this paper we conclude that (Sensys) 2003 Los Angeles 266-279.average delay is the best parameter for congestion detection [13] A. Woo and D. C. Culler, "A transmission control scheme for mediafor video and other multimedia traffic in WMSNs. Advantages access in sensor networks," In Proc. of the Annual Internationalof using this parameter are as follows: Conference on Mobile Computing and Networking (Mobicom) 2004 Rome 221-235. It has lower cost for congestion detection and besides has a [14] E. Paek, R. Govindan, "RCRT: Rate controlled reliable transport fordirect effect on quality of received video. Delay parameter not wireless sensor networks," In Proc. of the ACM Conference ononly uses local information but also it considers the whole Embedded Networked Sensor Systems (Sensys) 2007 Sydney status. Furthermore it is accurate in congestion [15] "The network simulator - ns-2," and it quickly detects congestion in network. The [16] J. Klaue, B. Rathke and A. Wolisz, "EvalVid a framework for videoother advantage is that it can be used in either sink or in transmission and quality evaluation," In Proc. Conference on Modeling,intermediate nodes. This leads to more flexibility in usage. Techniques and Tools for Computer Performance Evaluation 2003. quick congestion detection is aimed we may useintermediate nodes to detect and control congestion and ifreducing intermediate nodes overhead is favorable we can setsink in charge. One of the disadvantages of delay parameter is theoverhead of synchronization between nodes. We did notconsider this synchronization because this is not a major issue.Delay can be simulated with some heuristics and it canbecome independent of synchronization. So this is left forfuture work on the problem. REFERENCES[1] E. Gurses, O.B. Akan, "Multimedia communication in wireless sensor networks," Annales des Telecommunications/Annals of Telecommunications 60 (2005) 799–827.[2] I.F. Akyildiz, T. Melodia, K. Chowdhury, "A survey on wireless multimedia sensor networks," Computer Networks 51 (2007) 921–960. 1131