Dissertation/Thesis Abstract

Reconfiguration of multiprocessor systems with spares using genetic algorithms
by Bollineni, Venugopal, M.S., California State University, Long Beach, 2009, 50; 1472282
Abstract (Summary)

One of the most important issues in fault tolerance, computer architecture communication networks is the reconfiguration problems in faulty conditions. Most of the recently proposed reconfiguration schemes of replacing faulty nodes with spare nodes in parallel architectures address only for the small networks. This thesis is an investigation of the application of genetic algorithms (GAs) in reconfiguration of multiprocessor systems with spares. The problem is formulated as finding an optimal matching in a bipartite graph. The optimal matching is defined as the minimum sum of the distances of the links of the replacements of the faulty nodes by the spares nodes. The distance between the links is an important factor to provide a fast reconfiguration of the system in case of multiple failures.

The solution string may be encoded using the symbols representing the source nodes and the destination nodes. The fitness function is computed by adding the distances of the pairs of the source and the corresponding destination nodes. The parameters of the GAs are then applied through a number of iterations. Several operators are developed including the crossover operators. This technique can be used for large size of networks and can find the minimum sum in less period of time when compared to other techniques. Preliminary results of this study indicates that the GAs may be a cost-effective solution for the reconfiguration of the systems with spares.

Indexing (document details)
Advisor: Nguyen, Thinh
School: California State University, Long Beach
School Location: United States -- California
Source: MAI 48/02M, Masters Abstracts International
Subjects: Computer science
Publication Number: 1472282
ISBN: 978-1-109-47222-6
Copyright © 2021 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy