Time warp simulation using time scale decomposition

Abstract
In this paper we consider time scale decomposition as well as spatial decomposition to induce massive parallelism and reduce overhead in distributed discrete-event simulations. We confine our study to the Time Warp strategy and to systems where the durations of activities differ by several orders of magnitude (i.e., systems with fast and slow activities). We show that, for such systems, a large overhead due to rollbacks is encountered when spatial decomposition is used. Moreover, performance degrades as the difference increases between the rates of fast and slow events. Several initial experiments using queueing-network models were designed to evaluate the effectiveness of time scale decomposition in increasing the parallelism and reducing the overhead. These experiments were conducted on a distributed simulation testbed that was implemented on an 18-processor Multimax 320. The application of the above simulation techniques to stochastic Petri net models is illustrated using an example of performability analysis of a fault-tolerant distributed system.

This publication has 6 references indexed in Scilit: