On the effect and control of self-similar network traffic
Ki‐Hong Park · 1997
This paper presents a discussion of simulation-related issues arising in the study of self-similar network traffic with respect to its effect and control.Selfsimilar traffic has been shown to be an ubiquitous phenomenon arising in diverse networking contexts with potentially adverse effects on network performance.In many instances, an experimental or empirical approach needs to be taken to effectively evaluate the performance impact of sophisticated control algorithms acting at various layers in the protocol stack under self-similar traffic conditions, Simulating or experimentally implementing such environments is nontrivial due to the fact that, in general, the characteristics of the observed traffic is itself influenced by the actions of the control algorithms under study.To what degree self-similarity manifests itself in network traffic may depend on the properties of the protocols employed, and trace-based simulations that rely on traffic measurements to drive simulations fail to capture this dynamic aspect.We discuss an approach to evaluating control protocols under self-similar traffic conditions based on a simple, robust application-level causal mechanism of traffic self-similarity which is grounded in both empirical UNIX file system research and analytic traffic models involving certain renewal processes.We present a high-level discussion concentrating on simulation-related issues, with specific research results summarized or pointed to in the references.