1. IEEE 802.11 Rate Control Algorithms: Experimentation and Performance Evaluation in Infrastructure Mode Sourav Pal, Sumantra R. Kundu, Kalyan Basu and Sajal K. Das, the University of Texas at Arlington

2. Intro. IEEE 802.11 –multi-rate transmission capabilities by dynamically choosing the most appropriate modulation technique for the RSS. Rate Control Algorithm (RCA) –Open to the device manufacturer to improvise –Software/ Hardware approach Investigating the performance of software RCAs that interact with the PHY layer of the WNIC

3. Components WNIC –AR5212 from Atheros Driver –Madwifi 3 RCA algorithms –Onoe –Adaptive Multi Rate Retry (AMRR) –SampleRate

4. Goals Link layer –Qualitative and Quantitative Performance analysis of practical RCAs At the wireless link layer due to RSSI variation Application layer –Expose the impact of RCAs on application level throughput, packet inter-arrival time and jitter for heterogeneous traffic classes.

5. Rate Control Algorithms Onoe Credit Based RCA where credit is a function of number of successful and erroneous transmission/retransmission over a sampling period (1000ms). Implementation details: –10% or more needed retry, decrease credit else increase the credit –10 (0) credits, increase (decrease) data rate. –Once a data rate is failed, it will not attempt to select that until 10 seconds have elapsed Less sensitive to individual packet failure.

6. Rate Control Algorithms Adaptive Multi Rate Retry (AMRR) After 10 consecutive transmission successes –Transmits probe packets at higher rates to test the rate –Success  switch to higher rate AMRR employs Binary Exponential Backoff to adapt the probing threshold. –When the transmission of the probing packet fails, we switch back to the previous lower rate but we also multiply by two the probing threshold Reset success threshold to initial value if two consecutive transmission are failed

7. Rate Control Algorithms SampleRate Starts with the highest possible rate and then decreases till it can support that rate –Four successive failures Maintain the expected transmission time for each rate –Use the data rate with smallest expected transmission time Periodically transmits packets at rates higher than current transmission rates. –Computes the transmission time for every 10th packet which it sends in a different rate. Stale samples are removed based on a EWMA windowing mechanism.

8. Experiment Setup Disable RTS/CTS Upload a 8M file to content delivery server

9. Link layer Performance

12. Evaluation of application Layer Performance Radio range varied from 12 feet to 80 feet to ensure noticeable RSSI variation. Assume that mobility model is the same for each experiment. Ethereal is used to capture packet level statistics at both the AP and the laptop. tcptrace, an analysis tool written to extract statistics from ethereal dumps. Throughput presented is not end-to-end.

13. Application Layer Performance Skype, Kaffeine, firefox, sftp Average throughput (kbps)

14. Observations Significant difference between upstream and downstream bandwidth AMRR outperforms other two RCAs for non-real- time traffic Average Throughput for AMRR and Sample comparable SampleRate most suitable for Streaming; able to buffer and at distances of more than 70 feet where the rest of the RCAs fail.

15. Inter-packet Arrival Time OnoeAMRRSampleRate

16. Observations AMRR and SampleRate follow the RSSI and tries to move to a higher data rate as soon as wireless channel improves –Causing the transmission rate to fluctuate a lot Onoe is conservative –Switch to a bit-rate conservatively

17. Conclusion Performance Analysis of RCAs at link layer and application layer RCAs do not perform optimally on Low link condition –Probe, consecutive success/failure, long-term statistics The effect of RCA on application layer traffic Problems –Experiment should be carried with more clients –Impact of RTS/CTS has not yet been analyzed