Parallel simulstion of hierarchical, modular discrete event models

Abstract

The DEVS (Discrete Event Systems Specification) formalism specifies discrete event systems in a hierarchical, modular form. Within the formalism, state transition is specified by internal and external transition functions caused by scheduled internal events and randomly arrived external events, respectively. This thesis deals with the parallel simulation of DEVS models. The DEVS formalism and the associated abstarct simulator algorithm are implemented in a distributed environment and named as P-DEVSim++. For distributed simulation, two classes, Monitors and ModelIDs, are developed. Monitors acts as the distributed simulation kernel for DEVS simulation and ModelIDs supplies Monitors with mapping information and interfaces of all models and abstract simulators. In P-DEVSim++, internal events as well as external events of abstract simulators associated with DEVS models can be processed in parallel. For such processing, the PARINT algorithm, an optimistic simulation algorithm based on the Time Warp algorithm, is proposed. The proposed algorithm localizes rollbacks of a model within the model itself, not possible in the standard Time Warp approach. An advantage of such localization is simulation speedup. This thesis also devises mapping of DEVS models to utilize the parallelism of P-DEVSim++ in a hypercube multicomputer environment without multicasting. The coupling relationship between internal and external events is classified into four cases. For each case, a mapping algorithm to reduce the overall simulation finishing time is presented. Especially, task graphs are transformed to reduce the simultaneous external events synchronization overhead by the flattening and deepening methods. Also, when multiple internal events occur simultaneously and each internal event is associated with multiple external events, a hypercube is partitioned into a set of subcubes so that an internal event and the associated external events can be processed in the same subcube. The pe...