A filter-based self-similar trace synthesizer

Abstract

Recent empirical studies have shown that modern computer network traffic is much more appropriately modelled by long-range dependent self-similar processes than traditional short-range dependent processes such as Poisson. Thus, if its selfsimilar nature is not considered in the synthesis of experimental network data, incorrect performance assessments for network systems may result. This raises the need of a self-similar trace synthesizing algorithm with long-range dependence. In this paper, we propose and examine the feasibility of a filter-based method for the synthesis of self-similar network traces. The proposed approach can alleviate the problems encountered by conventional synthesizers, such as random midpoint displacement and Paxson's spectrum fitting, which cannot generate self-similar traces on the fly and may give negative numbers. Additionally, the extended range of self-similarity of the filtered approach can be easily managed by the filter truncation window; therefore, a trace that faithfully matches the measured behavior of true network traffic, where the self-similar nature only lasts beyond a certain range but disappears as the considered aggregated window is much further extended, can be generated.