Performance analysis of ATM ABR service under self-similar traffic in the presence of background VBR traffic

Recent empirical studies of traffic measurements of packet switched networks have demonstrated that actual network traffic is self-similar, or long range dependent, in nature. That is, the measured traffic is bursty over a wide range of time intervals. Furthermore, the emergence of high-speed network backbones demands the study of accurate models of aggregated traffic to assess network performance. This paper provides a method for generation of self-similar traffic, which can be used to drive network simulation models. The authors present the results of a simulation study of a two-node ATM network configuration that supports the ATM Forum`s Available Bit Rate (ABR) service. In this study, the authors compare the state of the queue at the source router at the edge of the ATM network under both Poisson and self-similar traffic loading. These findings indicate an order of magnitude increase in queue length for self-similar traffic loading as compared to Poisson loading. Moreover, when background VBR traffic is present, self-similar ABR traffic causes more congestion at the ATM switches than does Poisson traffic