AP-side WLAN Analytics

TL;DR

This paper introduces passive AP-side WLAN analytics frameworks, virtual speed test and uScope, enabling accurate estimation of client network performance metrics solely from AP observations, reducing the need for active measurements.

Contribution

It presents novel passive AP-side measurement frameworks for estimating client speed and uplink latency, validated through extensive real-world testing with high accuracy.

Findings

Estimation errors under 10% in over 1 million tests

Frameworks work on commodity hardware

Effective in diverse real-world environments

Abstract

Monitoring the network performance experienced by the end-user is crucial for managers of wireless networks as it can enable them to remotely modify the network parameters to improve the end-user experience. Unfortunately, for performance monitoring, managers are typically limited to the logs of the Access Points (APs) that they manage. This information does not directly capture factors that can hinder station (STA) side transmissions. Consequently, state-of-the-art methods to measure such metrics primarily involve active measurements. Unfortunately, such active measurements increase traffic load and if used regularly and for all the STAs can potentially disrupt user traffic, thereby worsening performance for other users in the network and draining the battery of mobile devices. This thesis enables passive AP-side network analytics. In the first part of the thesis, I present virtual speed test, a measurement based framework that enables an AP to estimate speed test results for any of its associated clients solely based on AP-side observables. Next, I present Uplink Latency Microscope (uScope), an AP-side framework for estimation of WLAN uplink latency for any of the associated STAs and decomposition into its constituent components. Similar to virtual speed test, uScope makes estimations solely based on passive AP-side observations. We implement both frameworks on a commodity hardware platform and conduct extensive field trials on a university campus and in a residential apartment complex. In over 1 million tests, the two proposed frameworks demonstrate an estimation accuracy with errors under 10%.