This paper presents some of the results of an empirical study of the performance of sequential and distributed discrete event simulations of closed, stochastic queueing networks. We have simulated four different network topologies under various loads. We present results for both conservative and optimistic synchronization methods.
In this paper we focus on the effects of lookahead and prediction on the communication and computation loads in distributed simulation. We have implemented four different versions of each simulation model--each with a different degree of lookahead. We present the lookahead ratio for each of these implementations and show how it is related to the average prediction per message. We argue that average prediction per message is an important parameter for characterizing the performance of distributed simulations. Average prediction per message is affected by network topology, load, lookahead, and the time synchronization method. We present empirical data relating average prediction per message to the communication and computation loads in our simulations.
Copyright 1989 by Computer Communications Networks Group, University of Waterloo.
Full text. BibTeX entry.