Performance-relevant Network Traffic Correlation

Correlation structure is an important metric to consider when modeling the performance of network traffic. Particularly, the presence of long-range dependence (LRD) in the input process may, under some circumstances, lead to poor queueing performance. In this paper, we first aim to characterize the conditions under which the presence of LRD is performance relevant. We define variations of an ON/OFF-type process that employ a truncated power-tail (TPT) distribution, and analyze their correlation structure in relation to queueing performance. Our analytic results show that the correlation structure in some cases is very sensitive to the presence of a background Poisson process, and that while other model variations exhibit LRD, it is only those with TPT-distributed ON times that queueing performance is poor. These results lead us to propose a procedure for extracting performancerelevant correlation properties, whose effectiveness we demonstrate via simulation experiments using synthetic and measured traffic.