1. Packet Loss Characterization in WiFi-based Long Distance Networks Authors : Anmol Sheth, Sergiu Nedevschi, Rabin Patra, Lakshminarayanan Subramanian [INFOCOM 2007] Reporter : 林緯彥

2. Motivation  Do some measurements for understanding how WiLD (WiFi-based Long Distance) networks perform in practice  Analyze the loss variability across time  Explore the solution and propose some methods to mitigate loss

3. Outline  Methodology  Loss Variability analysis  Remedies  Conclusion  Novelty & Strength  Weakness

4. Methodology  Measurements on a WiLD network testbed comprising of links in both rural and urban environment.  Use wireless channel emulator (Spirent 5500) to study each source of packet loss in isolation.

5. Methodology (cont.)  Testbed setup  802.11 a/b/g  CBR UDP traffic streams  Turn off MAC-layer ACKs and set the maximum retries limit to zero  Modify Atheros madwifi driver to pass up frames with CRC and PHY errors

6. Methodology (cont.)  Channel losses  External WiFi interference  External non-WiFi interference  Multipath interference  802.11 protocol-induced losses  Timeouts due to propagation delay  Breakdown of CSMA over long distances

7. Channel losses  External WiFi interference  External non-WiFi interference  Multipath interference

8. External WiFi interference  Any WiFi traffic that is not a part of the primary WiLD link is categorized as external WiFi interference experimentemulation

9. External WiFi interference  Effect of hidden terminals in WiLD networks

10. External WiFi interference  Effect of relative power and rate of external interference Emulation result

11. Channel losses  External WiFi interference  External non-WiFi interference  Multipath interference

12. External non-WiFi interference  Other devices that share the 2.4 GHz band.  microwave  cordless phone  Wide-band noise  Result : no significant correlation between noise and loss rate.

13. Channel losses  External WiFi interference  External non-WiFi interference  Multipath interference

14. Multipath interference  Comparing to mesh network deployment, there are two factors contributing to lower delay spreads in the WiLD networks.  long distance between two end hosts  line-of-sight deployment of the nodes Delays between a primary and secondary reflection at midway and quarter-way point.

15. 802.11 protocol-induced losses  Link layer recovery mechanism  Breakdown of CSMA

16. Link layer recovery mechanism  The 802.11 MAC uses a simple stop-and-wait protocol, when MAC ACKs are enabled, the sender has to wait for an ACK after each transmission, and this leads to decreasing channel utilization with increasing link distance. Emulation result

17. Breakdown of CSMA  On longer distance links, it is possible that the two nodes will begin transmission within the window defined by the propagation delay.  The throughput of the WiLD link degrades as the distance is increased.

18. Main Result

19. Loss Variability  Burst-Residual Separation  Burst  time-periods with sharp spikes in the loss rate  Residual  the losses that constantly occur in the underlying channel over time. P1P2 BurstResidual 01 Loss Variability distribution

20. Burst characteristic  Short burst  majority of the bursts to be short bursts that last for less than 0.3s  Long burst  a single long burst is followed by a string of other long bursts separated by short time- periods (in the order of a few seconds).

21. Residual characteristic  The residual loss distribution is stationary over hourly time scales  The residual loss rate on any link remains roughly constant over a few minutes even in the presence in short bursts during such periods.

22. Remedies  Frequency Channel Adaption  Rate Adaptation  Adaptive Forward Error Correction

23. Frequency Channel Adaption  The advantages of channel switching could be significant in presence of long or high- loss bursts Ch# Avg. Loss Number of switch 19.2- No adapt Lowest rate channel 6.840 Channel adapt Simulation result

24. Rate Adaptation  The increased transmission time of the frame increases the probability of a collision with the external traffic. Simulation result

25. Adaptive Forward Error Correction  At the end of each time slot the receiver informs the transmitter of the loss observed in the previous slot. Based on this link information, the transmitter adjusts the redundancy for the next round. Loss No FEC19.98 Adapt FEC 4.78 Simulation result

26. Conclusion  Most of the losses arise due to external WiFi interference on same and adjacent channels.  The loss due to external WiFi interference depends on the relative power level between the primary and external traffic, their channel separation, and the rate of external interference.

27. Novelty & Strength  The loss rate is strongly related to the amount of external traffic received on the same and adjacent channels in contrast to the omni-directional antennas used in the mesh networks deployment.  From the emulation traces they observed that almost 100% of the lost frames contained CRC errors.

28. Weakness  Switching the frequency channel could mitigate interference; however, it is not always possible to switch a frequency channel in a large scale deployment.  Most of the 802.11 radios have built in rate- adaptation algorithms which selects a lower rate with feasible encoding on experiencing high loss.

29. Thank you!