Medium Access Layer Performance Issues in Wireless Sensor Networks

Medium Access Layer Performance Issues in Wireless Sensor Networks Ilker S. Demirkol ilker@boun.edu.tr 13-June-08 CMPE, Boğaziçi University

Outline Background: WSN and its MAC Layer Properties Packet Traffic Modeling for Surveillance Wireless Sensor Networks Impact of a Realistic Packet Traffic Model on MAC Performance Contention Optimization for Contention-based MAC Protocols A Heuristic for Contention Optimization in WSN Video sensor networks and increasing their capabilities with contention optimization Improving Performance with a Better Slot Selection Method 2 / 66

WSN: Network of sensor nodes 1. Sense the environment 2. Process the sensed data 3. Communicate the processed data towards sink G. Boone, Sensors Magazine, 2004: Internet To browse stored information WSN To browse the reality 4 / 66

WSN: Many Application Areas Environmental applications e.g., habitat monitoring Health applications e.g., telemonitoring of human physiological data Military applications e.g., battlefield surveillance Home applications e.g., home automation... 5 / 66

1986 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 WSN publications in SCI-E (per year) 1600 1400 1200 1000 800 600 400 200 0 6 / 66

WSN Research No standardization in communication protocols. A challenging subject, since the sensor nodes are limited in: computational capacity: 4 Mhz CPU memory: 60 KB RAM communication bit rate: 20-250 Kbps power: battery generally infeasible to replace Primary objective is to achieve highest energy efficiency at each operation. 7 / 66

Energy Consumption Components Three energy consuming operations: Communication Sensing Computation To conserve energy: Shutdown the sensor (sleep mode) Duty cycle: Sleep-listen schedules e.g., duty cycle of 5 %: Work 5 unit time and sleep 95 unit time. 40 30 20 10 0 Energy Consumption (mj) 0.024 mj 8 / 66

Two Important MAC Layer Energy Wastes Idle listening Collision C XB A A good MAC protocol should reduce both types of energy wastes. The energy consumed per bit/packet is a crucial performance criterion. 9 / 66

Outline Background: WSN and its MAC Layer Properties Packet Traffic Modeling for Surveillance Wireless Sensor Networks Impact of a Realistic Packet Traffic Model on MAC Performance Contention Optimization for Contention-based MAC Protocols A Heuristic for Contention Optimization in WSN Video sensor networks and increasing their capabilities with contention optimization 10 / 66

Surveillance Wireless Sensor Networks (SWSN) SAFE Sink node UNSAFE 11 / 66

Packet Traffic Models for WSN Poisson, Periodic,... SPTM (SWSN Packet Traffic Model) Intrusion-initiated Bursty traffic Depends on: Sensing interval Intrusion detection model Sensor1 Sensor2 MAC Sensor3 12 / 66

Detection Models Binary Detection Model Probability of detection: 1 Elfes Detection Model Probability of detection: 1 0 d c Certain detection range sensor-target distance 0 d c d u sensor-target distance Certain detection range Uncertain detection range/sensing range Probabilistic d c d u d c detection degree = # of detecting sensors = 5 # of detecting sensors = detection degree = 3 13 / 66 # of sensors w detection prob.= coverage degree = 5

Consecutive Coverage Degrees Consecutive coverage degrees are not independent, i.e., can be repsented by: Markov process of order n = 2d v t T u s target speed sampling interval 14 / 66

SPTM Framework 3 2 5 3 We provided complete analytical model for: generation of first coverage degree relations of consecutive coverage degrees detection degree for a given coverage degree 15 / 66

SPTM Framework 16 / 66

SPTM Framework 17 / 66

Validation of SPTM Simulation parameters: p 1 0 0 20 distance Validation of coverage degree dependency: 18 / 66

Validation of SPTM Coverage degree distribution for all points: Coverage degree dependency example: Subsequent detection points of coverage degree 2: Different distributions! Simulations validates the analysis presented 19 / 66

Validation of SPTM Detection degree PMF for a given coverage degree. Correct PMF can be derived with analysis. 20 / 66

SPTM An application specific packet traffic model is derived by first defining a framework. The components of the framework is formulated analytically which are verified by simulations. The proposed PTM is adaptive: The detection parameter settings can be set according to the sensor type deployed The specifications of the SWSN application implemented can be incorporated such as the target area and density. 21 / 66

Packet Traffic Models for WSN Poisson, Periodic,... SPTM,... What is the importance of a realistic PTM? 23 / 66

Impact of Realistic PTM Performance of S-MAC is evaluated for three different PTMs: Periodic Binomial SPTM Simulation Parameters: 24 / 66

Load vs. Delay Similar average load results in very different average delays! 25 / 66

Effect of Scenario Parameters Investigated parameter: Target speed To use correct system parameters is also very crucial. 26 / 66

Limited Buffer Case Buffer size: 10 packets Packet drops start at lower loads with SPTM. 27 / 66

Impact of Using a Realistic Packet Traffic Model It is shown that, for surveillance applications, evaluating a MAC protocol with a packet traffic model other than SPTM may give misleading performance results. The results presented belong to performance evaluation for one-hop communication which will be accumulated in evaluations of the end-to-end performances. In fact, that will lead to higher divergence from the realistic results. 28 / 66

Energy and Delay Optimized Contention for Wireless Sensor Networks Contention-based MAC protocols Majority of the WSN MAC protocols are contention-based Optimize the energy consumed the delay incurred by the contentions of the nodes 30 / 66

Contention-based MAC protocols Medium access with contention slots: IEEE 802.11, IEEE 802.15.4 S-MAC, B-MAC, SCP-MAC, Z-MAC Packet tx... 3 1 2 time Procedure: Randomly select a slot Carrier sense till the slot you select If no tx before your slot, start tx. Contention Window (CW) What to do, in case of collisions? 31 / 66

Collisions ****Collision*** 2 3 1 2 Succ. packet tx... 1 3 Backoff: collision timeout Binary Exponential Backoff (BEB) IEEE 802.11, IEEE 802.15.4 Uniform Backoff (UB) S-MAC, Z-MAC, Sift, SCP-MAC, B-MAC time 32 / 66

The Significance of the Contention Window Size S-MAC defines the following settings: 63 contention slots Slot time: 20 bit tx. If only one node is contending : Expected contention delay (carrier sense duration): 600 bits tx The packet to be sent is 400 bits (S-MAC default)! 33 / 66

Trade-off Lower CW size: Mean carrier sense duration Probability of collision Higher CW size: Mean carrier sense duration Probability of collision 34 / 66

Impact of CW size on WSN performance We formulate the two crucial WSN objectives analytically: Overall energy consumed for contentions Delay incurred by contentions Parameters of the derived analytical formulas: Contention window size Number of contending nodes Collision timeout duration, communication speed, energy consumption values 35 / 66

Analysis vs Simulation Simulation results are average of 1000 runs. Parameters: Bandwidth: 20 Kbps Slot time: 1 ms The collision timeout: 15.15 ms (RTS+CTS+SIFS) 36 / 66

Contention Delay # of contending nodes = 5 : collision duration : carrier sense duration : contention delay 37 / 66

Contention Delay for Different Number of Contending Nodes The optimum CW size changes with the number of contending nodes. 38 / 66

Performance Comparison for Contention Delay SSCW propose an approximation for optimum Setting CW CW size. size to W* t reveals lower delays compared to other alternatives. SMAC defines fixed 63 slots. W t *: Delay optimizing CW size SSCW: IEEE Tran. on Wireless Comm., Nov 06, by Tian et al. 39 / 66

Energy Consumption Results # of contending nodes = 5 40 / 66

Performance Comparison for Energy Consumed W E *: Energy optimizing CW size 41 / 66

Energy-Delay Trade-off W E * vs. W t * Minimization of one metric can increase the value of the other 7.5% higher delay 8% higher energy 42 / 66

Optimized CW Size Conclusions Contention delay Derived analytically Delay optimizing CW size (W t *) Energy consumed for contention resolution Derived analytically Energy optimizing CW size (W E *) First work to present the optimization of CW size in WSN. The analytical work presented can be applied to all networks that defines uniform backoff. 43 / 66

ENCO: A Heuristic for CW Size Setting We need a method for incorporating CW size optimizations into a distributed implementation. Estimated Number of Contenders (ENCO) method for event-based applications average coverage degree ~ average of the number of contending nodes use optimized CW for average coverage degree throughout the network. 45 / 66

Estimated Number of Contenders (ENCO) Method Two types of event-driven applications: Random Event Locations (REL) application Correlated Event Locations (CEL) application SPTM is used for traffic generation of CEL application. 46 / 66

Simulation Parameters for Evaluation of the ENCO Method Parameter Value Range Number of nodes 20-200 Area 300 x 300 m 2 Sensing Range Target speed Sampling interval 50 m 10 m/s 1 second 1000 simulation runs with new deployments. At each run, 10 event points are logged. for CEL: 10 consecutive target detection points. 47 / 66

Simulation Results of the ENCO Method Sample REL Application: Sample CEL Application: W* ENCO W* E 48 / 66

ENCO A heuristic that utilize the contention optimization work is presented. It is defined for event-driven applications where the ENCO method is shown to successfully approximate the performance of the optimum contention window size. 49 / 66

Effect of CW Optimization on Video Sensor Networks Sensor nodes are equipped with video cameras: Controllable video frame rate Limited buffers (Motes: 60 KB RAM) Limited communication rate Mica2: 20 Kbps, MicaZ: 250 Kbps Low computational capacity Limited encoding alternatives 51 / 66

Simulation Parameters for VSN Performance Evaluation Network stack: S-MAC + GPSR Comm. range: 80 m 52 / 66

Simulation Results for ENCO in VSN Throughput: Latency: Increases throughput for all DC values Decreases latency 53 / 66

VSN Performance Improvement with ENCO In this work, we investigated a promising and challenging type of WSN, namely VSN. We show that with current hardware capabilities of the sensor nodes, VSNs can easily saturate even for low video rates. However, the ENCO method is shown to improve the VSN performance for both throughput and latency values which are important criteria for time-critical VSN applications such as border surveillance. 54 / 66

A Better Slot Selection Method? Mostly used slot selection: uniformly random IEEE 802.11, S-MAC, Z-MAC,... p* slot selection method is proposed to minimize the collision probability: (IEEE JSAC, 04) 56 / 66

Is It Really Better? Collision Probabilities (N=5): Average Contention Delays (N=5): 57 / 66

Which CW Size to Compare? For W* t : For individual W* s: Uniformly random slot selection results in better performance with W* t! 58 / 66

Choice of Slot Selection Method We showed that although p* results in fewer collisions, its producing a better overall performance depends on the contention window size used. The uniformly random slot selection performance is shown to be better than the p* method if it is used with the CW size of W* t that is derived analytically in this thesis. 59 / 66

Conclusions (1) The performance studies of MAC protocols are generally performed with either periodic or Poisson data traffic. However, in reality, the data traffic generated in WSNs depends on the application. We developed an analytical packet traffic model for surveillance wireless sensor networks which utilize a parametric and probabilistic sensor detection model. This model can be used to generate the time-based packet traffic load generated by an intrusion to a target area. Moreover, the analytical model derived can be used for performance evaluations of MAC protocols for SWSNs. By doing detailed simulations, we showed that the performance results of a communication protocol depend on the packet traffic model employed. Using a different model than SPTM can overestimate the performance of a MAC protocol. 60 / 66

Conclusions (2) Another work presented in this thesis is the energy and delay optimization of contentions for contention-window based MAC protocols. The importance of the contention window size setting on the energy consumption and the delay incurred is depicted both analytically and by simulations. Separate optimum contention window size setting methods are presented for the energy consumption and the incurred contention delay. It is shown that the performance of contention-based MAC protocols can be significantly improved using the optimum contention window sizes. The presented analysis and optimizations are also applicable to other wireless networks that define a fixed contention window size and uniformly random contention. 61 / 66

Conclusions (3) For practical implementation of the CW size optimization in event-triggered WSN, a heuristic namely, the ENCO method, is proposed. The simulations on two different type of event-triggered WSN applications show that the ENCO method achieves close performance results to the optimum contention window size results. In the simulations of ENCO for correlated event location scenario, the information of the number of nodes detecting the events is generated with the proposed SWSN packet traffic model. As a future work, the performance of the ENCO method can be increased with spatio-temporal information such as the history of number of contending nodes. 62 / 66

Conclusions (4) We also investigated the improvements in the capabilities of VSNs by using the contention optimization methods and the ENCO heuristic proposed. It is shown that increasing the quality of video capture frame rates results in a drop in the video quality received at the sink even for low rates with currently available hardware and encoding technology. The network performance results of VSN with the ENCO method show that this method can extend the capabilities of VSN by both decreasing the end-to-end delay and increasing the average number of frames received at the sink, i.e., the system throughput at the same time. As a future work, the ENCO method can be improved for QoS-aware WSN where different priority classes can use different CW sizes. 63 / 66

Conclusions (5) The CW size studies presented in this thesis base on the uniformly random slot selection method which is the most widely used slot selection method. However, an alternative method was proposed which minimizes the collision probability. This method is compared to the uniformly random method for the contention delay metric. It is found that depending on the CW size used, the slot selection method that resuls in better performance values changes. However, it is shown that the uniformly random slot assignment performs better when used with the optimized CW size. We state that the best slot seletion method investigation should be done in cooperation with the optimum contention window size for the method. As a future work, this joint optimization problem can be investigated. 64 / 66

Publications Journal Papers: I. Demirkol, F. Alagöz, H. Deliç and C. Ersoy, Wireless sensor networks for intrusion detection: packet traffic modeling, IEEE Communications Letters, vol. 10, no.1, pp. 22-24, January 2006. I. Demirkol, C. Ersoy and F. Alagöz, MAC protocols for wireless sensor networks: a survey, IEEE Communications Magazine, vol. 44, no. 4, pp. 115-121, April 2006. I. Demirkol, F. Alagöz, H. Deliç and C. Ersoy, The impact of a realistic packet traffic model on the performance of surveillance wireless sensor networks, Accepted to Elsevier Computer Networks, 2008. I. Demirkol and C. Ersoy, Energy and delay optimized contention for wireless sensor networks, under revision, Elsevier Computer Networks, 2008. I. Demirkol, E. Onur and C. Ersoy, Wake-up receivers for wireless sensor networks: Benefits and Challenges, under revision, IEEE Wireless Communications, 2008. I. Demirkol and C. Ersoy, Does minimizing the collision probability reveal the best performance? In preparation, 2008. I. Demirkol, A. Ozgovde, and C. Ersoy, Improving latency and throughput by intelligent contention window size adjustment for video sensor networks, In preparation, 2008. International Conference Papers: A. Ozgovde, I. Demirkol and C. Ersoy, Effect of sleep schedule and frame rate on the capabilities of video sensor networks, International Symposium on Wireless Pervasive Computing - ISWPC 08, Greece, May 2008. Y. Durmus, A. Ozgovde, I. Demirkol and C. Ersoy, Exploring the effect of the network parameters of video sensor networks, Accepted to International Symposium on Computer Networks, 2008. I. Demirkol, Evaluation of cluster-based cross layer protocols for wireless sensor networks, Accepted to IEEE INFOCOM 2006 Student Workshop, 2006. I. Demirkol, Evaluation of cluster-based cross layer protocols for wireless sensor networks, Accepted to ICN 06. National Conference Papers: A. Ozgovde, I. Demirkol and C. Ersoy, Uyuma çizelgesi ve çerçeve hızının görüntülü algılayıcı ağların başarımına etkisi, Akademik Bilişim, Çanakkale, 2008. I. Demirkol, K. Basol, O. B. Orhan and S. Sevinc, Sorgu-tabanlı telsiz algılayıcı ağları sınama ortamı çalışmaları, 65 / 66 Proceedings of Bilgi Teknolojileri Kongresi, pp. 449-454, Denizli, Turkey, 9-11 February 2006.

Comments/Questions? 66 / 66