Achieving Low Delay & High Rate in 802.11ac Edge Networks

Abstract

Provision of connections with low end-to-end latency is one of the most challenging requirements in 5G. In most use cases the target is for < 100ms latency, while for some applications it is < 10ms. In part, this reflects the fact that low latency is already coming to the fore in network services, but the requirement for low latency also reflects the needs of next generation applications such as augmented reality, virtual reality and the tactile internet.

In this thesis we analyze the end-to-end latency in an edge network where an 802.11ac wireless hop is the bottleneck and queueing delay at the AP is the main source of latency. We demonstrate that queueing delay is coupled to the aggregation level in 802.11ac WLANs and that we can manage the delay by controlling the aggregation level. We implement this algorithm with a simple feedback loop on Linux using MAC timestamps. We also propose and implement a machine learning technique to infer aggregation level from kernel timestamps on Android OS where we do not have access to MAC timestamps. We demonstrate that the aggregation-based rate control policy selects a rate between that of Cubic and BBR. Importantly, the end-to-end one-way delay is more than 20 times lower than that with Cubic and BBR while it induces very few losses. We also propose a passive technique using logistic regression to detect the location of the path bottleneck, i.e. whether the bottleneck is the backhaul link or the wireless hop, and show how measurement of the aggregation level can be used for this purpose. We show that this approach has more than 90% accuracy across a range of different network configurations.