Abstract
This paper considers the suitability of SPED, a synchronous parallel discrete event simulator, for the study of message passing networks. The simulation algorithm is described, and its potential performance is assessed showing that, under some simplifying assumptions, SPED might offer speedups directly proportional to the number of processors used in the simulation. An implementation of SPED in a distributed memory parallel system is used to study a model of an interconnection network for a multicomputer. Experiments show that SPED performs nearly as expected, as long as the event density imposed on the LPs is above a certain threshold. If this is not the case, the overhead due to synchronization plus communication dominates the execution time, and the achieved speedups are not as good. Some ways to improve the performance of SPED are proposed: a method to reduce the number of messages interchanged during the simulation, and a new algorithm for synchronous PDES, called PTD-NB (Parallel Time Driven- No Barriers), which reduces the synchronization overhead by removing barrier operations and can be easily implemented in multicomputer systems without support for global synchronization operations.
Keywords
Parallel discrete event simulation, synchronous PDES, multicomputer, networks, performance analysis, barrier synchronization
Date of this Version
December 1995