R/s-analysis of networking traffic for revealing long-term and short-term dependence r/s-analysis of networking traffic for revealinglong-term and short-term dependence

Abstract

Recent analyses of Internet traffic have shown that traffic volume fluctuations in wide-area and local-area networks are characterized by self-similarity, or long-range dependency. Self-similarity is a scale invariant property under timescale translation, i.e. it yields the existence of clustering and bursty characteristics in the flow over wide time scales. Thus, self-similar traffic causes larger queuing delays than the estimation by Poissonian traffic. Selfsimilarity is relatively a new concept in the computer networking community. It is very different from past views that considered computer network traffic as a Poisson process. In the case of self-similarity, burstness is present in all time-scales and does not tend to pure white Gaussian noise.