On the Interdependence of Congestion and Contention in Wireless Sensor Networks Mehmet C. Vuran Vehbi C. Gungor Özgür B. Akan School of Electrical & Computer Engineering Georgia Institute of Technology Atlanta, GA 30332 {mcvuran, gungor}@ece.gatech.edu Electrical & Electronics Engineering Department Middle East Technical University 06531, Ankara, Turkey akan@eee.metu.edu.tr Outline Wireless Sensor Networks (WSN) Congestion and contention in WSN Related Work Goals Evaluation Environment Results Conclusions Wireless Sensor Networks Internet, Satellite, UAV Sink Sink Several thousand nodes Distance of tens of feet Densities as high as 20 nodes/m2 Task Manager I.F.Akyildiz, W.Su, Y. Sankarasubramaniam, E. Cayirci, “Wireless Sensor Networks: A Survey”, Computer Networks Journal, March 2002. I.F.Akyildiz, M.C. Vuran, O. B. Akan, W. Su, “Wireless Sensor Networks: A Survey REVISITED” Computer Networks Journal, 2005. Wireless Sensor Networks (WSN) Characterized by the collaborative information transmission of densely deployed nodes High density leads to Local contention Network-wide congestion In fact, the level of local contention and the network congestion are closely coupled due to the multi-hop nature of sensor networks Network Congestion Network congestion leads to waste of communication resources leads to waste of energy resources hampers event detection reliability at the sink The WSN architecture employs unique sources for congestion Communication in a shared wireless medium Multi-hop nature of WSN Limited buffer size Main Sources for Congestion Channel Contention and Interference Contention occurs between different flows different packets of a flow Outgoing channel capacity becomes time variant High density exacerbates the impact of contention Number of Event Sources Higher number of event sources improve event detection efficiency Closely located source nodes increase contention Increased number of flows increase congestion Main Sources for Congestion (2) Packet Collisions Packet drops due to collisions may indicate lower congestion level Reporting Rate Increasing reporting rate causes network congestion even if local contention is minimized Many-to-one Nature Event communication between multiple sources and single sink causes bottleneck around the sink A comprehensive analysis of network congestion and local contention is required for WSN Related Work In [1], channel load information is incorporated into congestion detection and control mechanisms. [2] proposes transmission control scheme for use at the MAC layer. In [3], congestion detection is performed through buffer occupancy measurements. In [4], the backoff window of each node is linked to its local congestion state. It has been advocated in [5] that MAC layer support is beneficial in congestion detection and control algorithms. [1] C. Y. Wan, et.al., “CODA: Congestion Detection and Avoidance in Sensor Networks,” in Proc. ACM SENSYS 2003, November 2003. [2] A. Woo, et.al., “A Transmission Control Scheme for Media Access in Sensor Networks,” in Proc. ACM MOBICOM 2001, pp.221-235, 2001. [3] O. B. Akan and I. F. Akyildiz, “ESRT: Event-to-Sink Reliable Transport for Wireless Sensor Networks,” to appear in IEEE/ACM Trans. Networking, October 2005. [4] I. Aad, et.al., “Differentiation Mechanisms for IEEE 802.11,” in Proc. IEEE INFOCOM 2001, pp. 209-218, April 2001. [5] B. Hull, et.al., “Techniques for Mitigating Congestion in Sensor Networks,” in Proc. ACM SENSYS 2004, November 2004. Related Work (2) Cross-layer approaches in congestion detection and control is necessary in WSN There is a close coupling between local contention and network-wide congestion The interdependence of congestion and contention are yet to be studied Goals In this work, we investigate the interactions between contention resolution and congestion control mechanisms What are the consequences of independent operations of local contention resolution and end-to-end congestion control mechanisms? What is the effect of local retransmissions? What are the effects of network parameters such as buffer sizes of the sensors, number of sources and contention window size? Can cross layer interaction be performed by preserving the modularity of layered design or are cross-layer designs required? Evaluation Environment and Performance Metrics ns-2 simulations in a 100x100m2 sensor field One node selected as sink Nodes in an event area send information to the sink Performance Metrics Event Reliability (Rev) Number of Collisions MAC Layer Errors Buffer Overflows End-to-end Latency Energy Efficiency Number of Sources Event radius values non-congested region rthlow transition region rthhigh congested region 20m, 30m, 40m As reporting rate is increased reliability drops significantly Increasing number of sources, i.e., event radius, degrades reliability A common shape is observed for reliability Number of Sources (2) Close correlation between MAC layer errors and buffer overflows Buffer overflows start to build up as MAC layer errors saturate The maximum value of MAC layer error percentage occurs at rthlow For higher number of sources, congestion occurs at lower reporting rate Buffer Size Buffer size values 5, 50, 100, 250 Change in buffer size has minimal effect on reliability Buffer Size (2) Increasing buffer length increases percentage of MAC layer errors Small buffer sizes lead to lower latency If end-to-end latency is important, lower buffer sizes lead to acceptable reliability Since contention dominates, smaller buffer sizes are actually beneficial in WSN MAC Layer Retransmissions Retransmission limit values 4, 7, 10 Decreasing local reliability affects overall reliability rthlow occurs at lower values for decreased Rtxmax Increasing Rtxmax further have minimal effect on reliability MAC Layer Retransmissions (2) Local reliability level affects MAC layer errors In the congested region, end-to-end latency increases significantly Local reliability mechanism has converse effect on end-to-end latency Latency saturates in congested region and local reliability level affects the saturation reporting rate value Contention Window Average contention window values for source and router nodes Source nodes are located close Increasing reporting rate increases contention Contention occurs mainly in the vicinity of source nodes Adjusting initial contention window size, CWmin, may affect network performance Contention Window (2) Adjusting buffer size and CWmin leads to higher reliability In non-congested region, lower CWmin size is better As the reporting rate is increased, increasing CWmin improves reliability by 10% Adaptive contention window size adjustments lead to efficient results Reasons for Packet Drops Distribution of packet drops In non-congested region, packet drops are due to MAC and routing layers As reporting rate is increased, MAC layer errors saturate and buffer overflows dominate Adaptive reliability mechanisms are required considering traffic load Energy Efficiency Energy consumption increases with reporting rate in non-congested and transition regions Energy consumption saturates in the congested region Number of sources significantly effect energy consumption Energy Efficiency (2) Energy consumption is not significantly affected by buffer size or Rtxmax The effects of these parameters on other performance metrics enable energy-aware, adaptive protocols to be implemented Conclusions The interdependence between local contention and network-wide congestion is investigated Higher event resolution vs. higher contention Increasing number of sources improves event reliability Higher contention degrades network performance since sources are closely located Small buffer sizes may be beneficial For low reliability, low latency demanding applications, smaller buffer size leads to more efficient performance Local reliability vs. End-to-end reliability Higher reporting rate can be supported by local reliability In addition to local reliability, end-to-end congestion and reliability mechanisms required Conclusions (2) Traffic-aware contention window size The knowledge of reporting rate enables initial contention window size adjustments The effect of buffer size change can be given by contention window size adjustments Adaptive cross-layer reliability mechanism required Packet drop distribution changes dynamically Reliability mechanisms need to adopt to sources of drops Energy efficient adjustments are possible Energy consumption is minimally affected by buffer size and retransmission limit adjustments Local interactions directly affect overall network performance