The notions of the critical path of events and critical time of an event are key concepts in analyzing the performance of a parallel discrete event simulation. The highest critical time of any event in a simulation is a lower bound on the time it takes to execute a simulation using any conservative simulation mechanism, and is also a lower bound on the time taken by some optimistic methods. However, at least one optimistic mechanismis able to beat the critical path bound in a nontrivial way. In this paper we make a systematic study of the meaning of the critical path in parallel simulation, and describe criteria that determine when a simulation is bounded by its length and when it is not. We show (again) that no conservative mechanism can beat the critical path, but that at least four known optimistic mechanisms are all capable of supercritical speedup. We give performance data for the JPL Time Warp Operating System showing two specially constructed applications using different method...
David R. Jefferson, Peter L. Reiher