Topology-Aware Resource Adaptation to Alleviate Congestion in Sensor Networks

1. Topology-Aware Resource Adaptation to Alleviate Congestion in Sensor Networks Jaewon Kang, Yanyong Zhang, BadriNath IEEE transactions on Parallel and Distributed Systems Rutger university

2. Outline • Introduction • Problem Formulation • Capacity Analysis Model • Topology-Aware Resource Adaptation Scheme • Performance Evaluation • Conclusion and Remarks

3. Introduction

4. Introduction Before events occur, sensor node report data at a lower rate to save energy This is called dormant state When these events are detected, a high reporting rate is necessary to generate sufficient data to accurately depict the phenomena of interest TARA scheme activates appropriate nodes whose radio if off to form a new topology that has just enough capacity to handle the increased traffic Congestion is likely to occur because the data rates may exceed the capacity available from the currently active nodes

5. Problem Formulation

6. We have identified 3 typical hot spot within WSN

7. Problem Formulation Source Hot Spot As soon as an event takes place, these sources are likely to be within each other’s radio range. Can be eliminated by allowing only a small number of nodes to report to the sink. This will not degrade network services because these nodes are likely to report highly correlated observations due to geographic proximity If all these source nodes start sending packets at a high rate to the sink, a hot spot will quickly form, and a large number of packets will be dropped around the event spot.

8. Problem Formulation Sink Hot Spot When event occurred, sink nodes(and the nodes around them) handle a high traffic volume. Hence the batteries of these nodes around sinks will be drained quickly One way of alleviating sink hot spot is to deploy multiple sinks that are uniformly scattered across the sensor field and then balance the traffic among them

9. Problem Formulation Intersection Hot Spot The presence of multiple sources and multiple sinks results in more than one flow intersecting with each other. Due to the traffic merge at the intersection nodes, they can also become hot spots. Intersection hot spots are far more challenging because it is very difficult to predict the intersection points due to the dynamic nature of sensor networks. Hence can’t be avoided at deployment time, but demand runtime counter-measures. This paper will focus on alleviating intersection hot spots

10. Problem Formulation Traffic Control Resource Control • Resource control schemes seek to satisfy the fidelity level requirement of each flow, even during congestion, by assigning additional resources to the flow without taking resources away from other flows. The goal of traffic control is to tune the offered load of all flows to approach the optimal point to ensure that the resource is fully utilized.

11. Problem Formultion • Topology-Aware Resource Control • Naïve scheme: activate all the nodes and create multiple paths • Blindly scheme: activate a random number of nodes that are outside the congested area to detour packets. But there is no guarantee that the now topology can offer larger capacity than existing configuration • The meet the fidelity and energy requirements, an efficient resource control scheme should consider traffic rate, congestion level, and most important of all, network topology

12. Capacity Analysis Model

13. Capacity Analysis Model • The capacity of a given topology • Maximum throughput(packet delivery rate) that can be observed by the sinks • If there were no interference between links • The capacity of a topology would be the same as the maximum throughput achievable by unlimited unidirectional transmission • The interference between links, makes the overall throughput much smaller than the one-hop capacity • The objective of TARA’s capacity analysis model is to capture the degree of interference of a given topology

14. Capacity Analysis Model B T C D T Main idea behind TARA’s capacity analysis model is to map the problem of capacity estimation into a suitable graph-coloring problem T J I 2T H T G T Suppose sink receives m packets every n time frames. The capacity of this topology is calculated as m/n *Cmax CD CD HI HI Thus, the problem is reduced to proper coloring problem. Which means two adjacent vertex can’t draw the same color we construct the spatial interference graph , where the vertex denotes the corresponding wireless transmission to calculate m/n ( capacity fraction ) BC BC IJ IJ DI 5 colors are needed to color the vertices, and sink j receive 2 packets every 5 time frames. Thus the capacity fraction of this topology is 2/5 DI GH GH Edges between two vertices indicate that these two transmission are within each other's interference range IJ IJ

15. Capacity Analysis Model • Finding the minimum number of colors for a graph is NPC, however by using theorem, we can obtain an upper bound for colorability • If G is a simple with largest vertex degree d, than G is (d+1) colorable. • If G is a simple connected graph and not a complete graph, and if the largest vertex degree of G is (d>3), then G is d-colorable.

16. Capacity Analysis Model To solve our problem of estimating capacity fraction, we propose a heuristic solution HI BC GH CD We construct spatial reuse graph, which is the complement of the spatial interference graph IJ IJ DI Sort all vertices in ascending order of their degrees {CD},{DI},{HI},{IJ},{IJ},{GH},{BC} Start from the first vertex in the list and find the largest complete sub-graph comprise a concurrent transmission set , which include all the links that can transmit within the same time frame The number of concurrent sets is the minimum number of time frames needed to deliver a packet from each source to the sink

17. Capacity Analysis Model • The time complexity of our model is greatly affected by the number of nodes in the topology. • As a result, it may hard to apply our model over a large topology • Solution:take the viewpoint that the throughput of a topology is limited by the throughput of the bottleneck links • Therefore, only needs to calculate the capacity fraction for the portion of the network topology that contains the bottleneck links, referred to as bottleneck zone

18. Capacity Analysis Model Thus, we can focus on the appropriate bottleneck zone, which is much smaller than entire topology A B C T T D T T I J K L M 10T 2T 10T 2T 2T 2T T H G T E F T T Extract the bottleneck zone from a topology involves two steps identifying the bottleneck link and identifying all the links that interfere with it We introduce the term congestion sum of a link, congestion sum = that link’s traffic volume + the traffic volumes of all the links that cannot transmit concurrently with this link

19. Capacity Analysis Model • We validate our model against both simulation and actual experimental results by studying how to increase capacity for string, merging, crossing topologies • For each topology scenario, we have measured the capacity by using the following methods • Mote Experimentation • NS-2 simulation • Our Model

20. Capacity Analysis Model-String 2 2 3 3 1 1 • String topology is linear • We study how the capacity of a string topology varies with different hop counts between the source and sink • Our analytical model has the following capacity fractions: (l indicates path length) • if l = 1, capacity fraction = 1 • If l = 2, capacity fraction = 1/2 • If l≧3, capacity fraction = 1/3

21. Capacity Analysis Model-String Lesson 1 : minimizing the path length in a string topology does not increase the capacity if the resulting topology has a path length of more than two hops The result indicate that even though a string topology has a large hop count, it can provide a certain level of channel capacity Lesson 2: if the node whose incoming traffic volume is less than Cmin experiences congestion due to interference with other flows, the congestion can be eliminated by rerouting the incoming traffic onto the noninterfered path

22. Capacity Analysis Model-Merging Can be characterized by n+1 parameters: l1,…,lnand h, where li is the path length for the ith flow, and h is the number of hops between the merging point and the sink • Merging topology • Our analysis model derives the fellowing capacity fractions: • If h = 0, capacity fractions = n/(n+1) • If h = 1, capacity fractions = n/(2n+1) • If h≧2, capacity fractions = n/3n = 1/3 A B C T T D T T I J K L M 2T 2T 2T 2T T H G T E F T T

23. Capacity Analysis Model-Merging Lession3 :The capacity of a merging topology can be increased by moving the merging point within two hops away from sink.

24. Capacity Analysis Model-Crossing A crossing topology of multiple flows that have distinct sinks. Unlike the merging case, moving the crossing point, will not increase the capacity. One, may want to have multiple paths for either of the flows and split the traffic of tat flow onto these paths, as shown

25. Capacity Analysis Model-Crossing Lesson 4: to increase the capacity of crossing topology, at least one flow should have multiple paths and split its traffic onto these paths Scene 1 Scene 4 Scene 3 Scene 2

26. Topology-Aware Resource Adaptation Scheme

27. Build upon Capacity Analysis model

28. Adapt resources based on the congestion level

29. Topology-Aware Resource Adaptation Scheme Main idea: increase resources during crisis states Framework of TARA Issues: Congestion node detection Traffic Distributor Traffic Merger Detour Path Traffic Distribution Congestion Node Detection: Buffer occupancy Channel loading Hot spot node Traffic Distributor: Keep track each neighbors incoming packets Select the upstream neighbor that injects the most packets and send upstream control packet to that neighbor If that neighbor is also congested, repeats Distributor Merger Detour path

30. Topology-Aware Resource Adaptation Scheme Main idea: increase resources during crisis states Framework of TARA Traffic Merger: to locate merger, distributor traces downstream by sending a downstream control packet to the sink that the most traffic is destined for Merger should be located on the routing path with a low congestion level The choice of merger is dependent on the topology of the intersection zone, which include all nodes the two intersecting flows have in common Hot spot node Distributor Merger Detour path

31. Topology-Aware Resource Adaptation Scheme Issues when finding merger Braided intersection zone is formed when two flows with different sinks interfere with or without sharing nodes and do not cross

32. Topology-Aware Resource Adaptation Scheme Main idea: increase resources during crisis states Framework of TARA • Detour path: • Merger locally flooding the REQ packet including TTL field toward distributor • Discards all REQ if congestion • Discards all REQ if it is already on original routing path • Decrement TTL before forwarding and discard if TTL < 0 • Nodes Keep track largest TTL it has seen. Drop REQ whose TTL is lower • When REQ reach distributor a candidate detour path is established • Choose the largest one to be the path Hot spot node Distributor Merger Detour path

33. Topology-Aware Resource Adaptation Scheme Main idea: increase resources during crisis states Framework of TARA Traffic Distribution: Require distributor checks each packet’s dest before routing it Each detour path has a sink, and if a packet’s dest does not match the detour path sink, distributor will only send that packet to the original Among the packets have matching sink, we adopt the weight fair-share scheduling Hot spot node Distributor Merger Detour path

34. Performance Evaluation

35. Performance Evaluation • Performance metrics • Fidelity indexF/F0 • Total energy consumptionEtotal= • Bit energy consumptionratio of the total energy consumption with respect to the total number of bit successfully delivered to sink • Tool • NS-2

36. Performance Evaluation Each packet 100 byte long, outgoing buffer 10 packets 81 nodes, 160 square m2 communication range 30m,interference 50 m

37. Performance Evaluation • Compare 5 strategies • No congestion control • Traffic control • Deliver backpressure message to the upstream nodes to reduce the traffic load • Ideal resource control • Optimal offline resource control algorithm • Topology unaware resource control • TARA

38. Performance Evaluation-Merging

39. Performance Evaluation-Merging Fidelity index Total energy consumption

40. Performance Evaluation-Merging Bit energy consumption

41. Performance Evaluation-Crossing

42. Performance Evaluation-Crossing Fidelity index Total energy consumption

43. Performance Evaluation-Crossing Bit energy consumption

44. ConclusionRemarks

45. Conclusion and Remarks • TARA advantages • Energy efficient • Distributed • Topology aware • Issues • Upper watermark decision • Interval of crisis

46. BYE BYE