A Practical Approach for Multi-Scale Performance Analysis of Internet Traffic

摘要

Recent studies suggest that long-memory property in nowadays internet traffic shows a complicated scaling behavior: unlike the unchanged Hurst parameter of a self-similar process, scaling exponents of such processes change over time or scale ranges. In this paper, we propose a practical method, which uses the simple FBM process, to model long-range dependent traffic with changing scaling exponents over time scales. By using multiple self-similar processes to capture the statistical properties in different time scales, queueing performance of the multi-scale natured traffic can be presented by a composed performance curve produced by each self-similar process. Simulation results using traffic data measured from a backbone network is used to show the efficiency of our method.