The effectiveness of throughput sampling for capacity management: a queueing approach

For effective capacity management in access networks, it is essential to have a good insight in the service quality perceived by the users. As users share the service capacity available, one would want to know how the achieved per-user throughput fluctuates over time. In this paper we present a novel method that assesses the per-user throughput performance on the basis of throughput measurements at equidistant points in time. Our method relies on a queueing-theoretic framework, and allows us to explicitly quantify various statistics concerning the minimum per-user throughput obtained in a sample period, given the measured per-user throughput at the end points of that sample period, as well as the measured utilization during the interval. In an extensive numerical study we show the impact of important system parameters on these statistics. In addition, using illustrative examples, we demonstrate how the developed method can be used in practice for capacity management, with a specific focus on applications in cable access networks, an application for which our approach is particularly suitable.