On the accuracy of the effective bandwidth regenerative estimation

We consider the effective bandwidth (EB) estimation in the networks with the regenerative input. Based on the Lindley's-type recursion for the workload process in discrete and continuous time, we apply a large deviation approach to construct a regenerative estimate of the required EB in the nodes of tandem queuing networks. Simulation shows that the regenerative estimator overestimates the EB, while the standard batch means estimator implies underestimation. We study the accuracy of the regenerative estimation and, in particular, calculate the magnitude of the overestimation using either analytical expression (when available), or Monte-Carlo simulation. It is shown that for the considered models the overestimation is not significant, and thus regenerative estimation is relevant to evaluate the required capacities of the components of highly reliable systems.