Share Email Print
cover

Proceedings Paper

An automated parallel simulation execution and analysis approach
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

State-of-the-art simulation computing requirements are continually approaching and then exceeding the performance capabilities of existing computers. This trend remains true even with huge yearly gains in processing power and general computing capabilities; simulation scope and fidelity often increases as well. Accordingly, simulation studies often expend days or weeks executing a single test case. Compounding the problem, stochastic models often require execution of each test case with multiple random number seeds to provide valid results. Many techniques have been developed to improve the performance of simulations without sacrificing model fidelity: optimistic simulation, distributed simulation, parallel multi-processing, and the use of supercomputers such as Beowulf clusters. An approach and prototype toolset has been developed that augments existing optimization techniques to improve multiple-execution timelines. This approach, similar in concept to the SETI @ home experiment, makes maximum use of unused licenses and computers, which can be geographically distributed. Using a publish/subscribe architecture, simulation executions are dispatched to distributed machines for execution. Simulation results are then processed, collated, and transferred to a single site for analysis.

Paper Details

Date Published: 13 August 2004
PDF: 9 pages
Proc. SPIE 5423, Enabling Technologies for Simulation Science VIII, (13 August 2004); doi: 10.1117/12.542657
Show Author Affiliations
Joel D. Dallaire, Science Applications International Corp. (United States)
David M. Green, Science Applications International Corp. (United States)
Jerome H. Reaper, Science Applications International Corp. (United States)


Published in SPIE Proceedings Vol. 5423:
Enabling Technologies for Simulation Science VIII
Dawn A. Trevisani; Alex F. Sisti, Editor(s)

© SPIE. Terms of Use
Back to Top