This research proposal seeks to use fractal or self-similar models to describe the long-range correlations observed in measurements of traffic from actual networks, and apply techniques to deal with fractal models, well-known in physics but hardly ever used in the area of network performance analysis, to obtain relevant insight into the performance of modern high-speed networks for their design, management, and control. The research seeks to quantify the impact of fractal traffic characteristics on network performance, building on the present experience with self-similar traffic models. First, the work will extend the understanding of simple single-server models achieved by the PIs and other authors. The effects of complications such as multiple service classes and finite buffer sizes, and the modifications needed when self similarity is only asymptotic (relevant for variable-bit-rate services) will be considered. Second, the effects of connecting a large number of nodes in a network will be investigated including issues such as the robustness of self similarity. Finally, the project will tackle the difficult problems of understanding the transformations induced on self similar traffic by network controls as well as the consequences of self similarity on the effectiveness of the controls themselves. The work will consist of constructing simple models for networks, testing their validity through numerical simulations, and comparing their behavior with experimental measurements from real networks. The behavior of these models will be studied through a combination of numerical and analytical techniques.
