Analysis of some fluid models and a queueing network analyzer for polling systems
Abstract
The dissertation is organized into two parts. In the first part we discuss a queueing network analyzer for analysis of networks of polling systems under the exhaustive and gated service disciplines. The analyzer approximates network flows via correlated batch Poisson processes. The computational experience we report using the analyzer shows expected waiting times going to infinity for selected queueing networks known to be unstable at sub-critical loads. This computational experience prompted work toward the second part of the thesis--stability analysis of queueing networks via fluid modeling. In the second part we provide a counter-example (and propose an alternative mathematical-programming-based solution) to a recent conjecture in the literature. We also establish the strong stability of two fluid service disciplines for strictly sub-critical networks, namely the excess-capacity processor-sharing fluid discipline and the head-of-line proportional processor-sharing fluid discipline. The latter results are of interest to establish the existence of universally-stable fluid disciplines.
Degree
Ph.D.
Advisors
O'Cinneide, Purdue University.
Subject Area
Operations research|Industrial engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.