Approximate models for central server systems with two job types

Abstract

The purpose of this paper is to provide a new approximation technique for describing the performance of a closed queueing network when there are two or more job types present, and the order of service (scheduling policy) of interest does not make system states Markovian in the number present at each server; “first-come, first-served” order is an example. The model also allows incorporation of non-exponential distributions to represent service time distributions. Our technique drastically reduces the size of the state space used to describe complex processes. Consequently, it facilitates calculation of state probabilities and system performance measures through use of numerical methods such as Gauss-Seidel iteration. It is also helpful in simulation studies.