Synchronization support for efficient parallel discrete event simulation

Abstract

As a case study on the applicability of the EDA (extended dataflow architecture) model, CE-graphs (concurrent event graphs) have been defined to achieve high performance of parallel discrete event simulations (PDES) on parallel computers. Two levels of communication and synchronization are devised to execute CE-graphs efficiently, namely coarse-grained or medium-grained token (message) passing and fine-grained variable operations. The CE-graph approach can expose the highest degree of parallelism among independent events, while most existing approaches for PDES can only exploit parallelism at the level of components of a real system. Furthermore, the CE-graph approach can handle pre-emption situations in a natural way, which can be a difficult task for other approaches.< >