Determining Intra-Flow Contention along Multihop Paths in Wireless Networks

1. Determining Intra-Flow Contention along Multihop Paths in Wireless Networks K. Sanzgiri, I. Chakeres, E. Belding-Royer Department of Computer Science University of California, Santa Barbara Kimaya Sanzgiri

2. Introduction • Admission control is essential for enabling QoS • To make admission control decision, network must • Determine available resources • Estimate resource consumption of new flow • We focus on estimation of bandwidth consumption of a flow Kimaya Sanzgiri

3. Challenges in Estimating Bandwidth Consumption • Wireless transmissions consume bandwidth at all nodes within carrier-sensing range (assuming CSMA-based medium access) Carrier-sensing range of node Y Z Nodes W, X and Z cannot transmit when node Y is transmitting Y W X Kimaya Sanzgiri

4. Challenges in Estimating Bandwidth Consumption • Nodes along a multihop path may lie within each other’s carrier-sensing range • This leads to intra-flow contention Carrier-sensing range of node Y Z Packets of flow F contend for medium access at nodes W, X and Y Y W X Flow F Kimaya Sanzgiri

5. Intra-Flow Contention • Due to intra-flow contention, bandwidth consumption of a flow at a node becomes a multiple of that requested by the application Carrier-sensing range of node Y Z Bandwidth consumed by flow F at nodes W, X, Y each is 3 times the single-hop bandwidth Y W X Flow F Kimaya Sanzgiri

6. Contention Count • Contention Count (CC) at a node = Intersection of (set of carrier-sensing neighbors) with (set of nodes on multihop path) + 1 • Bandwidth consumption of flow = CC x (single-hop bandwidth consumption) • To estimate bandwidth consumption, CC must be calculated Kimaya Sanzgiri

7. Examples of Contention Count Contention Count at X = 5 Contention Count at Y = 7 V A U B P T C Q X Y R D S E F Carrier-sensing range of node X Carrier-sensing range of node Y Kimaya Sanzgiri

8. Challenges in Determining Contention Count • Node cannot directly communicate with all carrier-sensing neighbors (CSN) • Previously used approaches: • High power transmissions • Reduced spatial reuse, higher energy consumption • Multihop transmissions • Inaccurate, higher overhead Kimaya Sanzgiri

9. Related Work • Intra-flow contention ignored by most • CACP [Yang et al. 2003] • First to correctly consider intra-flow contention • High-power broadcast to communicate with CSN at each hop during reply phase of route discovery Kimaya Sanzgiri

10. Our Contribution • Two new approaches to determine contention count • No high-powered transmissions • Key Idea: Use carrier-sensing information from regular-powered transmissions to derive information about CSN Kimaya Sanzgiri

11. Carrier-Sensing Information • Graph of received signal strength vs. time • Duration of transmissions from CSN can be sensed RSS X RxThresh Y CSThresh time tx ty Kimaya Sanzgiri

12. Effect of Collisions Colliding packets X and Y are sensed as a single packet of duration tz • Collisions affect packet duration measurements if neither signal is sufficiently stronger than the other RSS RxThresh X Y CSThresh time tz Kimaya Sanzgiri

13. Proposed Approaches • Two approaches proposed: • Pre-Reply Probe (PRP) • Route Request Tail (RRT) • Integrated with route discovery of reactive routing protocol (AODV) • Nodes record duration of all sensed transmissions • Duration used to infer packet length (assuming common data rate) Kimaya Sanzgiri

14. Pre-Reply Probe (PRP) • Prior to RREP, destination sends Pre-Reply Probe Message (PRPM) • Size of PRPM randomly selected by destination • Identifies unique transmission duration • PRPM forwarded to source along route • Transmission duration at each hop recorded by CSN of that hop • Source locally broadcasts PRPM Kimaya Sanzgiri

15. PRP Example PRPM of length L PRPM of length L PRPM of length L PRPM of length L P Q R S Recorded pkt lengths: L L L L L L L L L L Kimaya Sanzgiri

16. PRP Example (cont.) P Q R S Recorded pkt lengths: L L L L L L L L L L Kimaya Sanzgiri

17. PRP (cont.) • Next, destination sends RREP • RREP contains size of corresponding PRPM (L) • Intermediate nodes calculate CC when processing RREP • CC = (Number of transmissions sensed of duration L) + 1 Kimaya Sanzgiri

18. PRP Example (cont.) RREP P Q R S L L Recorded pkt lengths: L L L L L L L L Contention count: 3 4 4 3 Kimaya Sanzgiri

19. PRP Analysis • Alleviates many of the drawbacks of CACP • Main advantage is no high power transmissions • Drawbacks: • Additional control message • Delay before RREP • Possible errors in case of collisions or retransmissions Kimaya Sanzgiri

20. Route Request Tail (RRT) • Removes some drawbacks of PRP • Random-sized tail attached to RREQ • At each hop • Unique tail size generated • Results in unique RREQ size • Tail of previous hop replaced • RREQ sizes accumulated in RREQ packet Kimaya Sanzgiri

21. RREQ packet in RRT Regular RREQ contents RREQ size at hop 0 RREQ size at hop 1 ….. RREQ size at last hop Tail attached by last hop Kimaya Sanzgiri

22. RRT (cont.) • Nodes record sensed packet durations as in PRP • Destination includes list of RREQ sizes in RREP • Intermediate nodes check which packet sizes listed in the RREP were sensed • CC = (Number of sizes sensed) + 1 Kimaya Sanzgiri

23. RRT Analysis • Retains benefits of PRP • Removes extra control message and delay • Increases RREQ size • Greater byte overhead • Higher probability of collisions and errors in duration measurement Kimaya Sanzgiri

24. Performance Analysis • Comparison with CACP • Analytical comparison • Simulation-based (NS-2) evaluation • 50 nodes in a 1500m x 650m area • CBR traffic • Number of sessions varied to evaluate performance under different load conditions • No mobility Kimaya Sanzgiri

25. CC Error Kimaya Sanzgiri

26. CC Latency Kimaya Sanzgiri

27. Control Packets Sent Kimaya Sanzgiri

28. Control Packets Received Kimaya Sanzgiri

29. Conclusion • Carrier-sensing information, such as duration of sensed transmissions, can be used to infer information about CSN • PRP and RRT determine intra-flow contention with low error • Error outweighed by benefits • Lower energy consumption • Reduced network load • Faster response time Kimaya Sanzgiri

30. Thank You Questions/Comments? Kimaya Sanzgiri