On the effect of concurrent applications in bandwidth measurement speedometers

Nowadays, operators are facing an increasing demand to ensure the committed bandwidth in broadband access links, both from customers and regulatory bodies. Usually, customers employ speedometers (such as Speedtest) to evaluate the access link performance. Such speedometers run in the user’s PC, concurrently with many other applications which may generate cross-traffic and reduce the processing time to the remaining applications. In this paper we analyze the impact of cross-traffic from other applications and CPU and memory occupancy due to concurrent applications in the speedometer accuracy, for a broad range of file-download and packet-train-based measurement techniques. We conclude that the packet-train technique is highly immune to cross-traffic, CPU and memory occupancy, which makes it amenable to use in general-purpose PCs running concurrent software. However, an estimation of CPU, memory occupancy and cross-traffic from other applications improves the measurement accuracy. In this regard, we present the QoS-Poll measurement software, which estimates the occupancy of system resources, including cross-traffic from other applications, and defines a region of acceptance of the bandwidth measurement in terms of the previous system parameters. The results of a QoS-Poll measurement campaign (more than 75 households) are also reported, showing excellent measurement accuracy.