Isolating Physical PER for Smart Rate Selection in 802.11

1. Isolating Physical PER for Smart Rate Selection in 802.11 Malik Ahmad Yar Khan and Darryl VeitchARC Special Centre for Ultra-Broadband Information Networks (CUBIN)The University of Melbourne, AustraliaIEEE Infocom 2009(acceptance rate = 19.65%)

2. Outline • Introduction • The core idea • RCA modification • Fragmentation implementation • RTS/CTS implementation • Performance • Conclusion

3. Introduction • Rate selection algorithms in use in 802.11 suffer from a fundamental problem • Implicity or explicity make decisions bases on total Packet Error Rate (PER) • False rate decision lowers the system throughput

4. Introduction • Total PER = PPER + CPER • PPER: Physical PER • CPER: contention PER • It is often the case CPER dominates PPER • Poor latency, loss and throughput

5. PER under a ideal environment

6. Goal of the paper • Providing a novel way to isolate the physical error • Providing a modular way of isolating the contention error which any rate adaptation algorithm can be enhanced

7. Fragmentation • Only consider a packet is fragmented into 2 frames • The fragment 2 is sent out only after an SIFS after the final ACK • Provided there are no hidden terminals

8. Fragmentation • If the first fragment is lost, for nay reason, the second will not even be sent • Observation: the loss of a second fragment on the first transmission attempt • Physical error

9. RTS/CTS • Once the RTS-CTS exchange is completed, the data packet will have the channel reserved, and will only encounter error for physical reasons

10. Block reservation in 802.11e • An option in IEEE 802.11e • The second data packet of such blocks could be used to directly measure PPER

11. Main concept • Not to use these methods all the time in practice • Only trigger their use often enough to gather good statistics on PPER

12. Scatterplots of PER with no cross traffic

13. Scatterplots of PER with CT and fragmentation

14. Scatterplots of PER with CT and fragmentation (RTS/CTS on)

15. Rate dependence of IP throughput

17. Fragmentation implementation • Only descriptors of second fragments are passed to the RCAs • A mechanism is needed to trigger the fragmentation of selected packets • Maintains a simple ON/OFF variable

18. RTS/CTS implementation • RTS/CTS protection is switched on by a per-packet decision • Use Long Retry Count (LRC) to record the statistic of RTS/CST transmission

19. Performance • Two main metrics • Weighted mean rate summaries • IP throughput • All experiments are run in IEEE 802.11a in testbed • Full rate set:{6,9,12,18,24,36,48,54} • Packet size: 1500 bytes • 10 stations if no further explanation

20. The wireless testbed • St[1-13] are stations, Sniff[1-2] are sniffers

21. Two implementation protocols • SampleRate • A MIT master thesis • Implemented on MadWifi • AMRR • A multi-rate algorithm also implemented on MadWifi

22. Single user with different RCAs

23. TPER for single user case for AMRR

24. SampleRate with different packet sizes (1472B)

25. SampleRate with different packet sizes (716B)

26. UDP upstream system throughput

27. TCP upstream system throughput

28. Conclusion • The paper investigates two ways to isolate Physical PER • RTS/CTS • MAC level packet fragmentation • Higher channel rates were used when PPER was in fact low, even when contention is present

29. Thank you!!