Self-similar traffic and its implications for ATM network design

Self-similar (or fractal) stochastic processes have been proposed as more accurate models of certain categories of traffic (e.g., Ethernet traffic, variable-bit-rate video) which will be transported in ATM networks. Self-similar processes exhibit a long range dependence structure which is not the case for traditional models. The distinct differences between these two classes of models, have significant implications for performance prediction and network design. We describe simulation results using both synthetic self-similar processes and empirical video traces. Based on these simulation results, we analyze certain existing congestion control schemes and show that, although these schemes may be promising under traditional models, they face serious challenge under self-similar models