Robustness of analytical models for nonclassical discrete event systems

Analytical models based on queueing-network theory have been observed to work well for discrete event systems which violate many of the classical assumptions, such as exponential service time distributions. This has been explained recently by the Operational Analysis approach. Use of this approach however, also involves an assumption on the service times, which can be considered restrictive. Here we give results showing that system performance measures are surprisingly robust to violations of this assumption. This justifies the use of operational analysis for predicting the performance of nonclassical discrete event systems. We also define a class of systems called monotonic systems, and give some new results for such systems.