Efficient simulation of multiprocessors through finite state machines

Abstract

This paper introduces a new approach to the implementation of event-driven multiprocessor simulators. Cache and memory behaviour is modelled through finite state machines which use a very limited amount of storage rather than a full execution context (CPU registers, stack). The resulting simulator design is conceptually simple and clean. Furthermore, we make the point that finite state machines can be scheduled faster. Our performance figures show that simulation overhead is lower than for comparable multiprocessor simulators.