Layer 2 vs. layer 3 mechanisms for improving TCP performance in 802.11 wireless LANs

Access points in 802.11 wireless networks suffer from performance problems because of insufficient resources at layer 2 - the DCF (distributed control function) access method provides equal access probability to all devices in a wireless cell including the access point itself. Consequently, performance degrades and contention between uploads and downloads can lead to the familiar TCP unfairness problem. In this paper, we study the measured performance of mechanisms at different layers for improving TCP performance in 802.11 wireless LANs. At layer 2, we consider the AAP (asymmetric access point) solution that keeps low buffer occupancy at the access point. At layer 3, we consider LAS-ACK, an adaptation of the least attained service (LAS) policy for wireless LANs that aims at minimizing the average queue size by giving priority to the shortest connections. Using an experimental testbed, we demonstrate that AAP is a good solution especially for multimedia (delay and jitter sensitive) transfers as long as upload traffic is low. On the other hand, LAS-ACK is very efficient at minimizing the durations of most upload and download transfers as long as the distribution of flow sizes is skewed enough. The price to pay for combining LAS-ACK and AAP is the requirement of deploying LAS-ACK on all wireless stations and not only at the access point. The resulting solution actually combines the positive effects of both solutions as LAS-ACK is less sensible to the distribution of flow sizes, while multimedia flows benefit from the short queue size at the access point.