Parallel logic simulation on a distributed memory machine

Abstract

The paper reports on an efficient logic simulation system using the Time Warp mechanism, implemented on a large-scale multiprocessor (Multi-PSI). The system includes local message schedulers, an antimessage reduction mechanism and a load distribution scheme to enhance performance. In the authors' experiment, using 64 processors, 50-fold speedup and 100 K events/sec performance was obtained. The paper also reports on an empirical comparison between the Time Warp mechanism and two other mechanisms: a conservative approach and a synchronous approach. The comparison shows that the Time Warp mechanism is the most suitable for large-scale multiprocessors.<>