Generalized Sensitivity Analysis

Abstract

Chapters 3, 4, and 5 primarily discussed a straightforward approach to the problem of constructing sensitivity estimates of a DEDS performance from a single sample path. The principal issue dealt with is the IPA algorithm, the consistency of the IPA estimate, and some extensions of the applicability of the IPA algorithms to situations where it does not appear to be applicable at the first glance. Chapter 6 by way of the “cut-and-paste” idea begins to address the broad issue of computing the performance of a general DEDS,J(θ), at different values of the system parameter(s), θ, or the response surface exploration in the experimental design terminology. In this chapter, we would like to continue this viewpoint and treat PA not so much as a particular algorithm or a group of algorithms but as an approach or mind-set towards the efficient processing and analysis of DEDS performance and to view PA as the first step in the eventual construction of a theory of DEDS very much in parallel with the control theory for CVDS (see Appendix C). We submit that PA-like techniques complement extant simulation and queueing network methodologies. The first three sections of this chapter are basically independent from each other and can be read in any order. However, they mutually reinforce one another by providing many connections and share perspectives with each other and with other parts of this book.