an approach for solving the unfairness problem in wlans
play

An Approach for Solving the Unfairness Problem in WLANs Martin - PowerPoint PPT Presentation

An Approach for Solving the Unfairness Problem in WLANs Martin Heusse * , Yan Grunenberger * , Elena Lpez-Aguilera ** , Andrzej Duda * * LIG Grenoble Computer Laboratory ** Catalan Institute of Technology Outline WLAN unfairness problem


  1. An Approach for Solving the Unfairness Problem in WLANs Martin Heusse * , Yan Grunenberger * , Elena López-Aguilera ** , Andrzej Duda * * LIG Grenoble Computer Laboratory ** Catalan Institute of Technology

  2. Outline • WLAN unfairness problem • Idea of Asymmetric Access Point • Implementing AAP • Performance of Asymmetric Access Point • Conclusions 2

  3. Unfairness problem wired • TCP connections to mobile stations - downloads, uploads • Sporadic UDP traffic with real-time requirements (VoIP) • We assume that wireless LAN is the bottleneck 3

  4. DCF characteristics • Half-duplex operation - One frame at a time • Equal channel access opportunities for all contending entities - AP and any of N stations ° statistical share of 1/(N+1) - Independent of frame length/transmission speed • Exponential backoff - Short term unfairness for larger N - Increased unfairness if bad channel conditions 4

  5. TCP Delayed Acknowledgements 40 39 38 37 33 35 41 40 39 38 35 37 • An ACK is delayed until (timeout or) - k segments are received ( k = 2 typical) - k : number of data segments per ACK 5

  6. N uploads k =2 DATA segments ACKs N =3 wired • kN data segments at stations, N ACKs at AP • AP share needs to be N/(N + kN) , 1/3 for k=2 • If share of 1/(1+N) - Short buffer at AP: losses, but ACKs are cumulative - Large buffer at AP: longer RTT, limited by flow control 6

  7. N downloads ACK DATA segments N =3 wired k =2 • kN data segments at AP , N ACKs at stations • AP share needs to be kN/(N + kN) , 2/3 for k=2 • If share of 1/(1+N) - Short buffer at AP: loss, limited by congestion control - Large buffer at AP: longer RTT, limited by flow control 7

  8. DCF; 4 uploads, 1 download 25 20 Data transferred (MB) Uploads 15 10 Download 5 0 0 5 10 15 20 25 30 8 Time (s)

  9. DCF; ping RTT vs. 4 downloads 60 50 40 Frequency 30 20 10 0 0 10 20 30 40 50 60 9 RTT (ms)

  10. A symmetric AP Approach • Give more channel access opportunity to AP - Asymmetric Access Point benefits from k more share than all stations in cell (or kN than any station) - Corresponds to the worst case (N downloads) - Increases performance in mixed upload/ download scenario - keeps the AP buffer empty so that TCP connections become self-clocked by the destination (short RTT over the wireless part) 10

  11. A symmetric AP Approach N =3 wired k=2 11

  12. Implementing AAP • Stations - Operate according to Idle Sense - Adapt CW to load conditions in the cell by observing the mean number of idle slots • Asymmetric Access Point - Constant CW value, independent of N! - Derived for given k and 802.11 variant 12

  13. Optimal CW target n ̄ i 6.0 500 5.5 optimal CW 5.0 300 n_i n ̄ i converges quickly CW proportional 4.5 to N 4.0 100 3.5 0 0 10 20 30 40 50 0 10 20 30 40 50 N N n ̄ i : average number of idle slots between transmission attempts 13

  14. AAP • CW proportional to N • AP Access probability proportional to N ➡ AP CW divided by N compared to STA CW s ➡ AP CW is a constant 14

  15. Measurements • Implementation of Idle Sense - Intel PRO/Wireless 2200BG 802.11 a/b/g cards - Modified firmware, operational cards • AP - FreeBSD box - constant CW • Stations close to AP , good channel conditions, 802.11 a at 54 Mb/s 15

  16. DCF; 4 uploads, 1 download 25 20 Data transferred (MB) Uploads 15 10 Download 5 0 0 5 10 15 20 25 30 16 Time (s)

  17. AAP; 4 uploads, 1 download 25 20 Data transferred (MB) Download 15 10 Uploads 5 0 0 5 10 15 20 25 30 Time (s) 17

  18. DCF; 1 upload, 4 downloads 25 20 Upload Data transferred (MB) 15 10 Downloads 5 0 0 5 10 15 20 25 30 Time (s) 18

  19. AAP; 1 upload, 4 downloads 25 20 Data transferred (MB) Downloads 15 10 5 Upload 0 0 5 10 15 20 25 30 Time (s) 19

  20. DCF; ping RTT vs. 4 downloads 60 50 40 Frequency 30 20 10 0 0 10 20 30 40 50 60 20 RTT (ms)

  21. AAP; ping RTT vs. 4 downloads 60 50 40 Frequency 30 20 10 0 0 10 20 30 40 50 60 21 RTT (ms)

  22. Conclusions • Simple solution at MAC layer to the unfairness problem - Right shares of transmission opportunity - Correct operation of TCP over 802.11 - self- clocked flow control - Keeps empty buffer at AP - gives short delays - Always preference to download connections • Optimal solution in mixed upload/download scenarios requires upper layer modification - Proper scheduling at IP/TCP layer 22

Recommend


More recommend