Comparison of Control Variates for Queueing Network Simulation

摘要

The purpose of this thesis was to compare several control variates for queueing network simulation. The author's goal was to provide the simulation community with some guidance for selecting control variates that will lead to significant reductions in the variance of the estimated responses that do not introduce significant bias. Both internal and external control variates were examined. The measures for comparing them were the variance ratios obtained for each control variate against each of the two response variables and the coverage of their respective 95% confidence intervals. The two response variables selected for this research were the average sojourn time in the network and the probability that the number in the fourth queue exceeds twice the mean number in queue at steady-state. The internal controls included standardized routing controls and standardized work variables. The external controls included the average sojourn time in the network and the average number in queue at the fourth node. The external controls were further classified into two groups: analytic Jackson controls and analytic approximations. (kr)