This paper presents an empirical study of the performance of various approaches to distributed discrete event simulation on a message-passing multiprocessor. This study compares the performance of distributed simulation to a sequential simulation of exactly the same system.
Four simulation benchmarks are presented. Each is a closed queueing network with a different topology. Four implementations of each benchmark are described--each with a different lookahead. Each benchmark was run under three customer loads. Each benchmark was run using one, two, four, and eight processors. This paper presents data showing the relationships between speedup, number of processors, lookahead, and computation load.
Copyright 1990 by Pennsylvania State University.
Full text. Presentation slides. BibTeX entry.