(M,P,S)-an efficient background traffic model for wide-area network simulation

This paper presents a background traffic model, (M,P,S), designed for use in wide-area packet-switched network simulators. The approach is to model the aggregate traffic generated by a large campus network and then partition the aggregate traffic into substreams, one for each campus-level destination in the backbone network. We use packet traces collected at the network exchange point of a large campus network and the global Internet to motivate our model design and demonstrate that our approach generates realistic results. Key features of the proposed model include: (1) efficiency-the generation of only a single long-range dependent stream is required for each campus network, (2) simplicity-the statistical properties (i.e., arrival density and correlation structure) of the sample path are controlled by a few parameters, and (3) accuracy - the self-similarity and arrival distribution for both the aggregate traffic and its component substreams accurately reproduce those of the empirical traffic trace. The results presented can be used to develop WAN background load models and traffic generators, contribute to understanding the fractal behaviour of Internet backbone network utilization, and provide a benchmark to evaluate the accuracy of existing traffic models.