Finite timescale range of interest for self-similar traffic measurements, modelling and performance analysis

Abstract

Although the existence of self-similarity in network traffic has been widely recognized, there is considerable debate on the impact of selfsimilarity on traffic engineering and network performance, and whether or not self-similar model should be used for traffic modelling. This paper reviews some major research in the area and summarizes limitations in current theoretical performance analysis with self-similar input. It is pointed out that these performance analyses are insufficient to make a definitive conclusion on the long-range dependence effects. Furthermore, it is pointed out that in a real system, the timescale range of interest for traffic measurements, modelling and performance analysis is limited, which can be characterized by an engineering timescale range (ETR). Only traffic correlations within the ETR will affect performance and are important for traffic measurement, modelling and performance analysis. Factors contributing to the ETR are identified. Further research is proposed on quantitatively identifying the ETR, and traffic measurements and self-similar traffic modelling using Markov models.