Uniform traffic does not represent a realistic view of traffic patterns in real systems. Non-uniform traffic models better reflect the traffic patterns that need to be accommodated. Such traffic may cause the network performance to deteriorate to much lower levels than the ones predicted by uniform traffic analysis. The authors extend the queueing analysis for buffered networks by providing methods for analyzing the queueing behavior of switching networks under non-uniform traffic patterns. They focus on shared buffer switch elements because they have better performance than input or output buffered elements. The analytical method for performance evaluation is compared with simulation on the basis of accuracy, where the performance is measured in terms of maximum throughput and probability of cell loss.

Presents an extension of the classical methods used to evaluate blocking probability. This method is applicable to multirate circuit and fast packet/ATM switching systems. The analytical methods are presented and compared with simulation results using Benes networks as an example. Extensive simulation has been performed focusing on ways to reduce blocking to acceptable levels. The author found that the major factors affecting blocking are switching element size, speed advantage of the switch fabric, routing algorithms and connection bandwidth.>

A study is made of an extension of the classical methods used to evaluate blocking probability. This method is applicable to multirate circuit and fast packet/ATM switching systems. The analytical methods are presented and compared with simulation results for Benes networks. Extensive simulation has been performed examining blocking in Benes networks. These simulations focus on ways to reduce blocking to acceptable levels. It is found that the major factors affecting blocking are switching element size, speed advantage of the switch fabric, routing algorithms and connection bandwidth.>