Exploiting Timed Petri Net Properties for Distributed Simulation Partitioning

This work systematically develops a distributed simulation mechanism based on the optimistic strategy (Time Warp) to study timed transition Petri net (TTPN) models. The approach addresses the problem of partitioning the simulation model into logical processes to be run concurrently on individual processing nodes in a distributed memory multiprocessor system. We propose a partitioning such that several transitions together with all their input places are simulated by a single logical process, in order to avoid the substantial overhead induced by a distributed connict resolution protocol in a message passing environment. Subnets are constructed from the topologic description of the TTPN so that connict resolution always occurs internally to a logical process, thus involving no communication overhead. Additional aggregation of connict sets into larger logical simulation processes may increase the balancing of the distributed simulation without decreasing its inherent parallelism if some conditions are veriied on the TTPN model structure.