Exact Sampling in Machine Scheduling Problems

Abstract

Analysis of machine scheduling problem can get very complicated even for simple scheduling policies and simple arrival processes. The problem becomes even harder if the scheduler and the arrival process are complicated, or worse still, given to us as a black box. In such cases it is useful to obtain a typical state of the system which can then be used to deduce information about the performance of the system or to tune the parameters for either the scheduling rule or the arrival process. We consider two general scheduling problems and present an algorithm for extracting an exact sample from the stationary distribution of the system when the system forms an ergodic Markov chain. We assume no knowledge of the internals of the arrival process or the scheduler. Our algorithm assumes that the scheduler has a natural monotonic property, and that the job service times are geometric/exponential. We use the Coupling From The Past paradigm due to Propp and Wilson to obtain our result. In order to apply their general framework to our problems, we perform a careful coupling of the different states of the Markov chain.