Local and Cumulative Analysis of Self-similar Traffic Traces

Abstract

Internet traffic shows variability in all time scales, which in turn shows statistical self-similarity. This selfsimilar behaviour has significant implications for QoS since it increments the total delay and packet loss rate. Therefore, we need to test for the degree of selfsimilarity and use this information for control purposes. For achieving the above-mentioned, the use of traces consisting of several thousands of points and hours of measurement are used. However, there are not enough studies about the number of points required to get an accurate estimation of the Hurst exponent. In this article, we study the local and cumulative behaviour of many real and synthetic self-similar traces. This is done for trying to infer the number of points required for Hurst parameter estimation and for checking dependence of Hurst exponents. We show that local analysis presents self-similarity, and the Hurst exponent tends to be stable in the cumulative case.