Sensitivity analysis of discrete event systems is an area that has recently attracted the attention of a number of researchers. Two general approaches have been proposed, perturbation analysis and the likelihood ratio method, that provide efficient estimates of parameter sensitivities for analytically intractable systems. Infinitesimal perturbation analysis estimates in particular are very attractive because they are easy to obtain, have small computational requirements and low variance. However, their usefulness is limited by the fact that they are biased in many cases. The proposed research is to investigate the possibility of using compensator identities in combination with perturbation methods in order to derive unbiased, low variance sensitivity estimates. Initial success in deriving estimates for the second derivative of the mean response time and of occupancy probabilities for the GI/G/1 queue justify confidence in this approach. The results of this study will be directly applicable in all the cases where discrete event simulation is used for sensitivity analysis and optimization of real systems and more specifically in manufacturing systems and data communication networks.
