On the impact of time scales on tail behavior of long-range dependent Internet traffic

Conventionally, Internet traffic has been modeled using classical Poisson-based models. More recent studies have proposed fractal models such as fractional Brownian motion. However, due to its simplicity, fractional Brownian motion is only efficient for approximating the performance of a class of exactly self-similar traffic, whose correlation property can be described by a single Hurst parameter. In this paper, we examine the tail behavior of long-range dependent Internet traffic, which has a more general correlation property. We propose an analytical method by focusing on the impact of time scales on queueing performance. The properties of traffic data are extracted from traffic traces of real networks, such as a wide area backbone network and a LAN. Results produced by simulation using real traffic data are compared with analytical results obtained by our method.