Comparing the Breathing Time Buckets Algorithm and the Time Warp Operating System on a Transputer Architecture

Breathing Time Buckets and Time Warp are two different approaches to optimistic parallel discrete event simulation. To explore their relative performance we ran two simple simulations using transputer-based implementations of the two algorithms. To exercise the two implementations thoroughly, we varied the value of the following parameters for one of the simulations, a queuing network: the number of objects involved; the granularity of events; and the distance into the future that the new events are scheduled. Both simulators provide speedup. The Breathing Time Buckets algorithm can perform better than the Time Warp based simulator for most cases. However, this latter behaves more consistently as it is not affected by either the distance into the future that new events are scheduled, nor by the network topology used. On the other hand, the performance of Breathing Time Buckets is sensitive to both of these parameters.