On the performance of time warp

There are two major approaches to performing distributed discrete event simulation (DDES): the conservative approach and the optimistic approach (Time Warp). An important area of research within DDES is the performance evaluation of distributed simulation algorithms.We implement Time Warp on an iPSC/2 hypercube and we study the performance of this optimistic approach in the context of queueing networks. The main focus of our study is to determine how the service time means of the logical processes affect performance. We employ two different network topologies, mesh and distributed (ARPANET), and we experiment with a varying number of logical processes. Aggressive and lazy cancellation strategies are explored.As our results indicate, Time Warp achieves significant speedups when the relative difference of the service time means is not high. A small number of IPs and a high degree of asymmetry in the simulated topology have a negative impact on Time Warp's performance