Masters Theses

Date of Award

5-1994

Degree Type

Thesis

Degree Name

Master of Science

Major

Computer Science

Major Professor

Jeffery D. Case

Committee Members

Dunigan, Mutchler

Abstract

The software framework presented herein, RCalc, provides Artificial Intelligence re- searchers programming in the Common Lisp language with a multiple-instruction, multiple- data (MIMD) super-computer built from a collection of Unix1 workstations interconnected via a local area network (LAN) utilizing the Transmission Control Protocol/Internet Protocol (TCP/IP) protocol suite.

RCalc is most suitable for parallel algorithms adhering to the bag-of-work paradigm. A model of bag-of-work algorithms as implemented using RCalc-like systems is presented. It is used to analytically derive a predictor for the efficiency, E, of parallel algorithms fitting the model.

To apply the predictor, one must know certain characteristics of the hardware and software realizing the algorithm. A testing methodology that can determine these char- acteristics is developed, and it is applied to RCalc. The results of this application are used to improve the predictor of E, and the improved predictor is validated by applying it and the testing methodology to a Parallelized Simulated Annealing Solver of the Travelling Salesman Problem (PSA-TSP) implemented in RCalc.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS