Efficient distributed simulation

Abstract

Discrete-event systems are used to model a number of engineering applications ranging from performance analysis of large scale communication networks, computer-aided-design (CAD) of circuits to simulation of manufacturing systems. Except for a small set, these systems are analytically intractable and in addition prohibitive to evaluate numerically. Simulation of such complex systems is exceedingly slow to run (and also to develop). Therefore, the development of simulation speedup methods is of crucial importance in its potential of making feasible the evaluation of a vastly larger set of engineering systems.New analytical models are proposed for the analysis of asynchronous, distributed simulation of discrete-event systems. Vectored distributed simulation and a new rollback algorithm, Wolf, are examined in this framework and comparison with existing systems is attempted. A number of examples illustrate the advantages to such a realization of distributed simulation. Implementation of our algorithms on a distributed computing system confirm analytical results predicting a considerable reduction in error propagation, and an enhancement in forward computation.