Automatic Parallelization of Discrete Event Simulation Programs
Fujimoto, Richard M.
MetadataShow full item record
Contemporary simulation applications usually need to process many millions of events to obtain sufficiently accurate performance predictions for large, complex systems. While parallel computation offers a solution to this problem, developing parallel simulation code is very time-consuming and currently requires a high level of expertise. In addition, traditional approaches to automatic parallelization, as used in many parallelizing compilers, are not well-suited for discrete event simulations because these computations are very irregular and exhibit highly data dependent behavior. In this paper, we present a framework to which many existing sequential simulation languages can be mapped. Using a Time Warp like parallel execution mechanism, we show that it is possible to automatically translate DES applications written in these sequential simulation languages into equivalent parallel programs. A case study of the SIMSCRIPT II.5 language is described to illustrate the proposed parallelization method. Based on this approach, a parallelizing compiler for the SIMSCRIPT II.5 language was developed. Preliminary performance data using this compiler is presented based on simulations of parallelized discrete event simulation programs.
Showing items related by title, author, creator and subject.
Su, (Philip) Shin-Chen (Georgia Institute of Technology, 1992-12)
An approximate load balancing parallel hash join algorithm to handle data skew in a parallel data base system Geum, Seong (Georgia Institute of Technology, 1995-05)
Perumalla, Kalyan S.; Savelsbergh, M. W. P. (Mathieu Willem Paul); Ramachandran, Umakishore (Georgia Institute of Technology, 1997)This report documents a framework called PARINO that we have developed for solving large mixed integer programs (MIPs)in parallel. PARINO combines portability and efficiency with flexibility. It is written in C++, and is ...