Paper Details

Experimental studies of conservative distributed discrete-event simulation on transputer networks

Authors: Cal, W., Turner, Stephen J.

Abstract:

Computer-based discrete-event simulation has a relatively long history. Traditionally, it has been performed in a sequential manner: the event-list simulation mechanism ([1]) is a typical example. The idea of distributed simulation was proposed by Chandy in 1977 and is now being developed mainly along two directions - the conservative approach (deadlock avoidance ([2]) and deadlock recovery ([3])) and the optimistic approach (time warp ([4])).Distributed simulation explores the potential parallelism inherent in most simulation applications. Each physical process (PP) in the application is simulated by a logical process (LP) in the simulation model. Events in the physical system are simulated by message transmissions between IPs. Since many simulation applications contain a high degree of parallelism, simulation seems to be a natural candidate for parallel processing. But, the causality constraint of the simulation, that is, events simulated by an LP must have a nondecreasing simulation time, is not easily maintained by distributed processing. Many strategies have been proposed: however, experimental studies need to be conducted in order to discover how much speed-up is achieved with a distributed simulation as compared to sequential methods. Previous performance studies by other researchers ([5,6]) have mainly been carried out on shared-memory parallel processors. In this paper, a set of experimental results is presented, designed to evaluate the effectiveness of conservative distributed simulation strategies on message-passing parallel processors such as transputers.

Proceedings:

OUG-12: Tools and Techniques for Transputer Applications, Stephen J. Turner, 1990, pp 138 - 147 published by IOS Press, Amsterdam

This record in other formats: