Analysis of bandwidth measurement methodologies over WLAN systems

TL;DR

This paper analyzes how WLAN transmission dynamics affect bandwidth measurement tools, proposing an analytical model that reveals biases in delay distribution and clarifies the actual metrics these tools estimate.

Contribution

It introduces a new analytical model for WLAN bandwidth measurement that accounts for wireless-specific behaviors, improving understanding of measurement biases and metrics.

Findings

Delay distribution varies within probing sequences.

Existing tools estimate achievable throughput, not true bandwidth.

Model validated through experiments and simulations.

Abstract

WLAN devices have become a fundamental component of nowadays network deployments. However, even though traditional networking applications run mostly unchanged over wireless links, the actual interaction between these applications and the dynamics of wireless transmissions is not yet fully understood. An important example of such applications are bandwidth estimation tools. This area has become a mature research topic with well-developed results. Unfortunately recent studies have shown that the application of these results to WLAN links is not straightforward. The main reasons for this is that the assumptions taken to develop bandwidth measurements tools do not hold any longer in the presence of wireless links (e.g. non-FIFO scheduling). This paper builds from these observations and its main goal is to analyze the interaction between probe packets and WLAN transmissions in bandwidth estimation processes. The paper proposes an analytical model that better accounts for the particularities of WLAN links. The model is validated through extensive experimentation and simulation and reveals that (1) the distribution of the delay to transmit probing packets is not the same for the whole probing sequence, this biases the measurements process and (2) existing tools and techniques point at the achievable throughput rather than the available bandwidth or the capacity, as previously assumed.