An implementation of a technique for sharing variables in Time Warp

Abstract

For some systems, the use of shared variables is a natural means to model the desirable interactions in a parallel discrete event simulation. We propose an implementation of a technique which uses a full replication strategy to share variables in distributed memory systems using Time Warp. The implementation handles causal effects and tie breaks between accesses. We show that the technique can be implemented in three different ways: aggressive cancellation, lazy cancellation and optimised scheme of rollback in order to enhance performance. An evaluation of implementations of the technique is provided using an example. >