Conservative protocols for distributed discrete event simulation are strongly dependent on the ability to identify good lookaheads, that is, intervals in simulation time during which it can be guaranteed that no event-carrying message can be generated or received by a local processor (LP). The authors observe that in certain kinds of simulation problems (the authors particularly refer to colliding pucks and road traffic simulations), regions of LPs can lag behind the rest of the simulation, owing to reduced or vanishing lookaheads. To make the simulation robust in such situations, the authors propose a new method for accelerating the Chandy-Misra-Bryant Conservative Null Message Protocol. The acceleration approach identifies a lagging set of LPs and then runs a protocol between them that allows them to identify a time to which they can all safely advance. The protocol for cooperative acceleration is described, with all the possible state transitions. A proof of correctness is provided. Numerical results from actual simulations show the improvement that arises from the acceleration method.
In the absence of wider information, conservative protocols assume the worst case and hence sometimes end up with very small or vanishing lookaheads. The authors solve this problem by providing another layer of protocol that makes wider information available to the LPs, thus allowing them not to assume the worst case. It is clear that such a proposal will work. Two key questions do not seem to be addressed, though: How does the simulation identify a subset of LPs among which the protocol should be initiated? and When should the protocol be initiated? These questions must have been answered in the implementations that the authors have based their numerical results on, but the answers are not given in the paper. Further, it seems that the rows in Figure 7 (shouldn’t it be Table 7?) have been interchanged, as these numbers show CES performing better than ABCMB.