1. A Mini-survey of Dealing with Faults in Wireless Sensor Networks Qi Han

2. Motivation Current WSNs exhibit high loss rates In indoor environments, half of the links: 10% packet loss A third links: 30% packet loss At MAC layer, link-layer re-transmissions are unable to mask this loss Assuming packet loss rate is p, then the probability that a message is successfully received Across n hops is (1-p) n Understanding packet delivery performance In dense wireless sensor networks J. Zhao and R. Govindan, SenSys 2003 (Best Paper Award) Taming the underlying challenges of reliable multihop routing in sensor networks A. Woo, T. Tong and D. Culler, SenSys 2003

3. Strategies to deal with faults MAC layer Apply ARQ Network layer: Select high quality paths for data transmission Multi-path routing Braided Diffusion GRAB (Gradient Broadcast) Transport layer Downstream data delivery PSFQ, GARUDA Upstream data delivery TAG, ESRT

4. Braided Diffusion How to perform energy-efficient and robust dissemination of data from sources to sinks tradeoff between resilience and energy consumed Based on directed diffusion: Construct dissemination path from multiple sources to multiple sinks on- demand D. Ganesan, R. Govindan, S. Shenker, D. Estrin MobiHoc 2001 and MC2R 2002

5. Directed Diffusion a)Source periodically broadcasts events at a low rate b)Sink sends a reinforcement message to one of its neighbors c)The message is propagated to the source, hop by hop d)When a node on the reinforced path fails, the sink re-initiates reinforcement Drawback: A periodic low-rate flooding scheme notifies the sink and other nodes of available alternate paths --- Consumes energy

6. Disjoint Multipath Localized algorithm: -using local information - use two kinds of reinforcements

7. Braided Multipath - Alternate paths in a braid are partially disjoint from the primary path

8. Failure Models (used for evaluation) Isolated failures: capture independent node failures Patterned failures: capture geographically correlated failures

9. Gradient Broadcast The sink builds a cost field Cost at a node: minimum energy overhead to forward a packet from this node to the sink along the path The cost field gives the global direction towards the sink implicitly At each hop, only nodes that have costs smaller than the sender can forward the packet F. Ye, G. Zhong, S. Lu, L. X. Zhang ACM WINET 2005

10. Credit-based Forwarding Mesh Limit the ‘width’ of the forwarding mesh More than enough paths of decreasing cost exist A source assigns a credit to the packets it sends out Credit: An extra budget that can be used to send a packet to the sink along a path The amount of credit controls the redundancy of the mesh

11. Reliable Downstream Sensor Data Delivery Data flows from sink to sources for the purpose of control or management PSFQ Assumptions: message loss occurs due to the poor quality of wireless links Hop-by-hop recovery: node In-sequence forwarding GARUDA: reliable delivery To all sensors, To a sub-region, To minimal sensors to cover the sensing field To a certain percentage of the sensors

12. Reliable Upstream Sensor Data Delivery Data flows from sources to sink TAG (Tiny Aggregation): A node switches its parent in two cases: Each node monitors the quality of the link to each of its neighbors by tracking the proportion of packets received from each neighbor When a node observes that it has not heard from its parent for some fixed period of time, it assumes that its parent has failed

13. ESRT: Event-to-Sink Reliable Transport in WSN Event! A sensor node A sensor node that can sense the event Sink wants reliable event detection with minimum energy expenditure [Y. Sankarasubramaniam, O. B. Zkan, I. F. Akyildiz, ACM MobiHoc 2003] [Slides modified based on the class presentation of A. Abouzeid from RPI]

14. Problem Definition Motivating application: Reliable detection/estimation of event features based on the collective reports from a number of sensors, not on individual sensor reports The sink must decide on the event feature every  time units Definition The reliability of even feature is measured by the number of received data packets Observed event reliability r i Desired event reliability R Problem statement: (congestion solution) Model any increase in source information as a increase in the sensor reporting rate f To configure the reporting rate f of source nodes so as to achieve the required event detection reliability R at the sink with minimum resource utilization

15. Evaluation Environments -- to study the relationship between f and r ns-2 simulator 200 sensor nodes 100m x 100m area 40m transmission range 30 byte packets 65 packets buffer size 10 sec decision interval (τ)

16. Effect of varying sensor reporting rate f on the event reliability r network gets congested sooner with increasing number of source nodes r linearly increases with f until f=f max, then drops After f max, it is wavy with increasing n, the drop in r is more significant This confirms the need for a reliable transport solution with a congestion control mechanism

17. CongestedNot Congested Lower reliability than required Higher reliability than required OOR Five characteristic regions Goal: To stay in OOR where energy expenditure is optimal

18. Main Idea of ESRT Sink Based on current state Si, calculates a updated reporting frequency fi+1, broadcasts it to sensor ndoes Si {(NC,LR),(C,LR)}: aggressively update f to reliably track event ASAP (Primary objective: reliably detect event) Si{(NC,HR),(C,HR)}: decrease f conservatively (Secondary objective: conserve energy) Sensors Listen to the sink broadcast at the end of each decision interval and update f Deploy a local congestion detection support mechanism

19. ESRT Actions Network State Action (NC,LR)Multiplicatively increase f, Achieve required reliability ASAP OORf remains unchanged (NC,HR)Decrease f conservatively Cautiously reduce energy consumption while not compromising reliability (C,HR)Decrease f carefully but aggressively to (NC,HR) to relieve congestion Then, follow (NC,HR) behavior (C,LR)Decrease f exponentially to relieve congestion ASAP

20. Stability of ESRT ESRT converges to OOR from any of four initial states {(NC,LR), (NC,HR), (C,HR), (C,LR)} From (NC,HR), ESRT stays in the state until converges to OOR Convergence time depends on ε – smaller ε causes longer convergence time

21. Congestion Detection Congestion status is required at the sink to determine the network state Based on expectation of buffer overflow at sensor nodes During a single interval, f and n do not change much If pending congestion (b k +  b>B) is detected CN bit is set in event reports

22. From (NC,LR) Reaches OOR in two intervals

23. From (NC,HR) ESRT stays in (NC,HR) until reaching OOR in five intervals

24. (C,HR) to (NC,HR) then OOR

25. (C,LR) to (NC,LR) then OOR

26. Power savings from (NC,HR) Reporting rate gets reduced conservatively while maintaining reliability

27. What I like about the paper Collective reliability Individual sensor ID is not necessary Each source attaches event ID Biased implementation Almost entirely in sink

28. What I dislike about this paper Sink must broadcast the updated reporting frequency at high energy so that all sources can hear it Ongoing event transmission would be disrupted Regulating all sensors to have the same reporting rate may not work well with heterogeneous sensors Assuming that sensors report periodically may not be true for all applications Congestion in WSN not just caused by frequent sensor reporting

29. My class project - reliable upstream data delivery Acquisitional query How to interpret the answer A Is A based on a very incomplete subset What about the remaining sensors Query may specify its reliability requirements(how good it wants A to be) Percentage of sensors (e.g. A is based on reports from 80% of the sensors) Recall (answer-set/exact-set)

30. Quality Metrics Sensors of interest N Answer N yes Missing at most N-N yes -N no N no Answer N yes recall r=N yes /(N-N no )

31. Problem Description Given: 1 sink (where continuous queries are injected) and n sensors with constant failures in the sensor network Objective: minimize energy consumption s.t. r>=R, where r=N yes /(N-N no )

32. Issues to Address How to distribute r q from the sink to intermediate nodes ? Given the assigned r q ’, what should each intermediate node do after finishing the transmission of the reports from all children? Go to sleep immediately Re-transmit immediately Stay awake in case re-transmission is requeted

33. Our Approaches To come

34. Evaluation Methodology ns-2 simulator Comparison against Plain collection (CSMA/CA) Link layer acknowledgement (CSMA/CA + ack/re-transmission) Multi-path routing