Self-similarity and long range dependence on the internet: a second look at the evidence, origins and implications

Abstract

In this paper we critically reexamine some of the long standing beliefs regarding self-similarity and long range dependence (LRD) on the Internet. Power law tails have been conjectured to be a cause of LRD. In this paper, we reexamine the claims regarding heavy tails. We first examine the generative models for the heavy tail phenomena, both in terms of the fragility of some proposed mechanisms to modeling perturbations as well as the weak statistical evidence for the mechanisms. Next, we take a look at some of the implications of LRD in key performance aspects of Internet algorithms. Finally, we present an alternative model explaining the LRD phenomena of Internet traffic. We argue that the multiple time-scale nature of the generation of traffic and transport protocols make the observation of LRD inevitable.