Learning to Wait: Wi-Fi Contention Control using Load-based Predictions

Abstract

We propose and experimentally evaluate a novel method that dynamically changes the contention window of access points based on system load to improve performance in a dense Wi-Fi deployment. A key feature is that no MAC protocol changes, nor client side modifications are needed to deploy the solution. We show that setting an optimal contention window can lead to throughput and latency improvements up to 155%, and 50%, respectively. Furthermore, we devise an online learning method that efficiently finds the optimal contention window with minimal training data, and yields an average improvement in throughput of 53-55% during congested periods for a real traffic-volume workload replay in a Wi-Fi test-bed.