Recently, an approach using the discrete event paradigm for the simulation of continuous systems has been developed. This approach is based on the use of piecewise-linear approximations for the representation of continuous, time-varying quantities. The focus of this paper is to show how this new technique can be implemented on a multiprocessor. Our hypothesis is that the Time Warp algorithm is the natural protocol for parallel simulation using continuous, piecewise-linear signals.
We have implemented a prototype simulator and measured its performance on a multiprocessor using several benchmarks. On the basis of these benchmarks we show that the best performance is achieved when using a scheduling algorithm that gives priority to messages with lower timestamps and when the lazy-cancellation variant of Time Warp is used. In this case, the performance of our prototype simulator compares quite favourably with the estimates of optimum performance based on a critical path analysis of the sequential simulation. In fact, our results indicate that in certain cases the performance of the Time Warp algorithm using lazy cancellation can exceed the conservative optimum estimate obtained from the critical path analysis.
Copyright 1991 by Ablex Publishing Corporation.
Full text. BibTeX entry.