1. 1 Congestion Performance Improvement in Wireless Sensor Networks (WSN) Junjie Xiong, Michael Lyu, Kam-Wing Ng The Chinese University of Hong Kong March 7, 2012 2012 IEEE Aerospace Conference Bigsky, Montana

2. 2 Outline Motivation LIFO Queueing Mechanisms for Congested WSNs Multi-queue Mechanisms for Congested WSNs Performance Evaluation Conclusion

3. 3 Outline Motivation LIFO Queueing Mechanisms for Congested WSNs Multi-queue Mechanisms for Congested WSNs Performance Evaluation Conclusion

4. 4 WSN Introduction Application-oriented  Surveillance, e.g., space exploration  Target tracking Resource-constrained  E.g. Energy supply: 2 AA batteries  E.g. RAM: 10k bytes Subject to dynamic traffic changes Region Base station (BS) Sensor nodes Internet

5. 5 Motivation Serious problems happen in congested WSNs.  Delay degradation: Base station (BS) cannot receive data from the sensor nodes in time  Fairness degradation: BS fails to receive data from far-away nodes Current congestion-related methods  Congestion avoidance method  Congestion recovery method No method considers improving the congestion performance in the aspect of the queueing mechanism  We are the first to suggest using LIFO (Last-In, First-Out) to replace FIFO (First-In, First-Out) in congested WSNs  We also employ multi-queue mechanism

6. 6 Outline Motivation LIFO Queueing Mechanisms for Congested WSNs Multi-queue Mechanisms for Congested WSNs Performance Evaluation Conclusion

7. 7 LIFO Queueing Mechanisms for Congested WSNs Delay performance comparison of LIFO and FIFO. According to the well-established theory on a congested M/M/1 queue: LIFO achieves better delay performance than that of FIFO LIFO also achieves better delay performance than FIFO in congested WSNs.

8. 8 Fairness Definition In a time interval T, suppose the BS receives x i packets from the i-th node, and N is the total number of nodes in the WSN. Then the fairness index the WSN is defined as: If x i is equal to each other, then f(x) =1 (fairest).

9. 9 Fairness Performance Comparison of LIFO and FIFO (1) Setting:  Mean service time: T s = 3s  Mean arrival time: T i = 1s  Deadline of data freshness: T d = 30s.  The WSN is overloaded: T s /T i > 1 Fairness: LIFO is better. For FIFO, the nodes far away from the BS may never have data received by the BS. For LIFO, the nodes far away from the BS will have data received by the BS.

10. 10 Fairness Performance Comparison of LIFO and FIFO (2) T s = 3s, T i = 1s, T d = 30s.

11. 11 Fairness Performance Comparison of LIFO and FIFO (3) T s = 3s, T i = 1s, T d = 30s.

12. 12 Outline Motivation LIFO Queueing Mechanisms for Congested WSNs Multi-queue Mechanisms for Congested WSNs Performance Evaluation Conclusion

13. Size of trees:  Tree BS:10  Tree 7: 3  Tree 8: 6  Tree 9:1  Tree 14: 3 13 Multi-queue Mechanisms for Congested WSNs (1) V V/3 V/9 V/12 V/36 No logical multi- queue at each node

14. Multi-queue mechanism is adopted to improve fairness:  Each node maintains a queue for itself, and a queue for each of its child nodes.  Each node gives higher service capacity to packets from queues with larger length. 14 Multi-queue Mechanisms for Congested WSNs (2) V 3V/10 6V/10 V/10 3V/10 V/10 Maintain a logical multi-queue at each node

15. 15 Outline Motivation LIFO Queueing Mechanisms for Congested WSNs Multi-queue Mechanisms for Congested WSNs Performance Evaluation Conclusion

16. 16 Performance Evaluation – Delay (1) 196-node network Queue size: 30, data generation interval T i = 200s

17. 17 Performance Evaluation – Delay (2) Queue size: 30, deadline T d = 400s

18. 18 Performance Evaluation – Fairness (1) Queue size: 30, data generation interval T i = 200s

19. 19 Performance Evaluation – Fairness (2) Queue size: 30, deadline T d = 400s

20. 20 Outline Motivation LIFO Queueing Mechanisms for Congested WSNs Multi-queue Mechanisms for Congested WSNs Performance Evaluation Conclusion

21. 21 Conclusion We are the first to improve the delay and fairness performance of congested WSNs with queueing mechanisms: First, we employ LIFO instead of FIFO to improve both the delay and fairness performance. Second, we implement a logical multi-queue mechanism. Finally, our experiments show that the multi-queue- LIFO mechanism achieves better delay and fairness performance than the traditional FIFO.

22. 22 Q & A Thank you!