Self-adaptive logical processes: the probabilistic distributed simulation protocol

A probabilistic distributed discrete event simulation strategy is developed as a performance efficient compromise between the two classical approaches in parallel and distributed simulation, the conservative and the optimistic approach. It weakens the conservative “block until safe-to-process”-rule in a sense that if the time instant of the occurrence of an external event is in the time interval [s, t], it allows progressing simulation up until the forecasted next event instant tˆ(O), s⩽tˆ(O)⩽t, but further progression only with controlled probability. tˆ(O) is an estimate based on the arrival instant differences O=(δ₁,δ₂,...δ_n) observed during a time window by some logical (simulation) process. Compared to the optimistic strategy it prevents from propagating incorrect computations too far ahead into the simulated future, and thus avoids unnecessary communication overhead by breaking rollback cascades as early as possible. The arrival patterns observed in O are used to dynamically adapt the logical process´ synchronization behavior at runtime, to what is the best tradeoff among blocking and optimistically progressing with respect to the parallelism inherent to the simulation model