A goal of this paper is to explore different ways of implementing distributed simulation. Distributed simulation grew out of sequential simulation, and it is possible that the way we think about distributed simulation is unduly influenced by its sequential origins. To free ourselves from unnecessary restrictions on the way we design distributed simulations, in this paper we define the distributed simulation problem somewhat differently than in the literature. We propose the concepts of 'knowledge' and 'conditional knowledge' to help us obtain a general framework to reason about distributed simulation without too close a coupling with any specific implementation method. The framework appears helpful in designing new ways of distributed simulations. Empirical studies of distributed simulations report widely varying results: some studies report improvements in speed that are almost linearly proportional to the number of computers in the system, while other studies report that distributed simulation is even slower than sequential simulation. The framework proposed in this paper seems to help in explaining the wide difference observed in empirical studies. Using our framework, we attempt to suggest properties that efficient 'general- purpose' distributed discrete-event simulations must have. This paper assumes little prior knowledge of the literature on simulation or distributed systems. We hope that the paper will serve as a tutorial in addition to providing additional insight.
展开▼
