Scott Suhr

Date of Award

6-1991

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering

Department

Department of Electrical and Computer Engineering

First Advisor

Gary B. Lamont, PhD

Abstract

Government agencies and academic institutions are very interested programming for concurrent processing to cut computer processing time. However, many of the world's problems have already been coded for conventional serial computers. This research demonstrates the feasibility of modifying existing serial codes for execution in a concurrent processing environment. A electromagnetic scattering prediction code known as NECBSC is incrementally modified to incorporated various levels of concurrent computing. The data processed by the code are completely independent, providing an avenue for data decomposition of the process. Portions of the data set are processed on each mode and the results combined for final output. The final version of the code demonstrates a speedup of 3.59 on an eight mode iPSC/2, verses the serial benchmark on that machine. Speedup for the iPSC/860 is 2.51, lower (vs its baseline) because of the faster processor, but it's elapsed time is shorter by 23%. Significantly better efficiencies are achievable when a more complex situation is simulated due to the relatively constant volume of output/ communications. The success of this effort demonstrates that, at least for problems easily data-decomposed, the decomposition and implementation of existing serial codes for execution in a concurrent environment is both possible and profitable.

AFIT Designator

AFIT-GE-ENG-91J-04

DTIC Accession Number

ADA238802

Download

Share

COinS