Performance evaluation of pseudo self-similar traffic

Abstract

Several studies demonstrated that network traffic exhibits self-similarity and long-range dependence. Of the new methods for modeling self-similar traffic, the pseudo self-similar traffic model (PSST) (Robert and Le Boudec (1997)) is especially attractive since it is simple, has only two parameters to fit and is readily analyzed by classical queueing theory methods. In this paper we perform a performance evaluation of the PSST model by investigating its suitability to model the queueing behavior of self-similar traffic. We simulate a system fed by self-similar traffic traces with a single server and geometrically distributed service times. The observed steady state queue size distribution is compared to the one predicted by a PSST model fitted to the parameters of the traffic traces.