Date of Award

1995

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

J. Bush Jones

Abstract

Bush Jones published a series of papers providing sequential algorithms that are key to reconstructability analysis. These algorithms include the determination of unbiased reconstructions and a greedy algorithm for a generalization of the reconstruction problem. The implementation of these sequential algorithms provide scientists and mathematicians with the means of utilizing reconstructability analysis in systems modeling. The algorithms, however, are so computationally intensive that the system is limited to a very small set of variables. Many papers have been written applying reconstructability analysis and maximum entropy methods to various disciplines. Reconstructability analysis has the potential of dramatically impacting the scientific community, but the sequential algorithms leave the utilization of reconstructability analysis infeasible. The author has parallelized the reconstructability analysis algorithms developed by Jones, thereby, bridging the gap between theoretical application and feasible implementation. Since the goal of parallelization of these reconstructability analysis algorithms is to make them feasible to as many researchers as possible, a specific architecture is not assumed. It is assumed that the architecture employed is a multiple data architecture. That is, the architectural design needed for the implementation of these algorithms must have memory local to each processing element (PE). The parallel algorithms developed and presented here do not address the problems of communications between processors of particular architectures. These algorithms assume a reconfigurable bus system which is a bus system whose configuration can be dynamically altered thus allowing broadcasting and long-distance communications to be completed in constant time. It is noted that processor arrays with such reconfigurable bus systems have been designed. Frequently, parallel algorithms do not address the situation in which the number of values on which to operate is larger than the number of processors. However, since the purpose of the parallelization of these reconstructability analysis algorithms is to make them feasible for large structure systems, the parallelization given does address the situation in which the number of values on which to operate is larger than the number of processors available. Therefore, implementation of the algorithms involves simply incorporating the communication protocols between processors for the particular architecture employed.

Pages

124

Share

COinS