1. Reducing Congestion Effects in Wireless Networks by Multipath Routing Presented by Dian Zhang Lucian Popa, Costin Raiciu, University of California, Berkeley ICNP 2006

2. Outline  Problem statement and assumptions  BGR  congestion control mechanism  IPS  EPS  Performance evaluation  Conclusions

3. Overview a point-to-point communication in a large Wireless Network

4. Overview

5. Goal: Split flows and reduce congestion by having less traffic in a contention/interference area.  Improve overall throughput  Improve fairness

6. Multipath Routing-related work  We need alternate paths to avoid congested hotspots  Existing solutions not satisfactory  Way points  Simple  Course Grained  Trajectory Based Forwarding [Niculescu03]  Fine Grained  Complex and resource consuming

7. Biased Geographical Routing (BGR)  Idea  Insert a “bias” inside the packet as a measure of the deviation from the greedy path  Achieve different paths by using different biases bias = 0 small bias large bias

8. BGR Details Destination bias1  Bias = Angle  Route greedy towards a close by point projected at “bias” angle  Decrease bias at each step  Quadratic dependence on distance to destination  Stop decreasing when it reaches zero

9. BGR Details Destination bias2  Bias = Angle  Route greedy towards a close by point projected at “bias” angle  Decrease bias at each step  Quadratic dependence on distance to destination  Stop decreasing when it reaches zero

10. BGR Details Destination bias3  Bias = Angle  Route greedy towards a close by point projected at “bias” angle  Decrease bias at each step  Quadratic dependence on distance to destination  Stop decreasing when it reaches zero

11. Overview We propose two algorithms to deal with congestion 1. IPS - In-network Packet Scatter  Local algorithm  Lightweight – no per flow state  Suited for short flows or light congestion 2. EPS – End-to-end Packet Scatter  End to end – rate control, relies on receiver feedback  Suited for long flows and widespread congestion

12. IPS (In-network packet scatter) Learn about congested neighbors

13. IPS (In-network packet scatter) Deviate uniformly traffic to congested nodes on three paths

14. EPS (End-to-end packet scatter) Source Destination

15. EPS (End-to-end packet scatter) Source Destination Congested node

16. EPS (End-to-end packet scatter) Source feedback Destination

17. EPS (End-to-end packet scatter) Source Destination

18. EPS (End-to-end packet scatter) Source Central path is prioritized Destination Exterior paths are less aggressive than the central one

19. EPS (End-to-end packet scatter) Source Most traffic on least congested path Destination

20. IPS+EPS  The two algorithms can be used  Independently  Combined

21. ns2 setup  400 nodes grid  802.11 wireless  Random source-destination pairs  Success measured as received number of packets

22. Throughput vs Hop Count Increase on AIMD(%)

23. Throughput vs No. Transmissions

24. Testbed  Mirage sensor network testbed  ~100 nodes  But too “narrow” – interference on one side  Thus, our main goal was to estimate in practice potential for throughput increase

25. Testbed metodology  Throughput of vs

26. Testbed results RateLQI 80LQI 82LQI 84LQI 86 40 packets/s167117242 33 packets/s 120-1217-8 20 packets/s 93-3-2-12  % Increase in received packets for two BGR paths of 40 degree bias compared to single paths

27. Conclusions  BGR  An efficient and practical multipath algorithm for wireless networks with location information  IPS, EPS  Two mechanisms to increase fairness and throughput by multipath routing  Practical tests

28. Thank you! ICNP 2006