Generation and testing of self-similar traffic in ATM networks

Abstract

A number of findings from detailed studies of traffic measurements from different packet networks have brought up a surprising discrepancy between the traditional traffic modelling techniques and the actual network traffic. The studies have shown the actual network traffic to be statistically self-similar with significant implications for the design of future multi-service integrated networks. This new traffic feature can be effectively captured within fractal models like: fractional Brownian motion (fBm) and fractional ARIMA processes. Although these formal mathematical models provide an elegant solution to the modelling of the self-similar phenomena, an comprehensive queuing analysis of these models is still lacking. Therefore simulations with synthetic self-similar input traffic are essential for gaining better understanding of the queuing problems and some initial experience with the performance of the future networks. Consequently fast generation of long traces of self-similar processes becomes an important task. We use an fBm generation method called the successive random addition (SRA) algorithm and carry out a rigorous statistical analysis on the generated traces. Our results show that the traces are indeed self-similar, although the parameters obtained may slightly differ from their target values. Our conclusion is that for qualitative studies the SRA algorithm provides a very good traffic source, whereas for quantitative analysis some caution is recommended. We also mention some possible applications of the algorithm in performance-related network implementations.