Statistical Self-Similarity in Broadband Traffic: Results and Performance Implications

Abstract

In this paper, we utilize data traffic measurements from the Ethernet LAN at IIT, Delhi to demonstrate the self-similar nature of Broadband network traffic. This ‘scale-invariant’ or self-similar behavior, detected visually as through the use of statistical techniques, is drastically different form both conventional telephone traffic and from stochastic models for packet traffic traditionally considered in the literature. We discuss the far-reaching implications of fractal nature on the design, management and performance of high-bandwidth, cell-based networks. The failure of classical traffic models to capture self-similarity necessitates the development of new mathematical approaches that are simple, accurate and realistic for aggregate data traffic. Issues such as the above, which arise in the light of this new understanding of broadband traffic, form the crux of this paper.