Research On Parallel Simulation Engine And The Relevant Techniques

As the price of parallel architecture computer is decreasing and the research subjectsare being more complex, people turn to high performance computing for simulationof complex systems. This makes parallel discrete-event simulation(PDES) again a newresearch focus today. Parallel simulation engine bears great importance in PDES research.This dissertation does research on issues in parallel simulation engine and the relevanttechniques from three aspects: parallel simulation engine architecture, PDES performanceoptimization, and usability improvement.Firstly a four-layered parallel simulation engine architecture is proposed and PARSE,a prototype system is implemented. Each of the four layers: communication, kernel, RTIservice, and function extention layer, is studied in detail on functionality and the relevantPDES techiques, especially the kernel layer and function extention layer. In PARSE, twospecial logical processes are implemented for real-time parallel simulation and interactionwith HLA federations.Using PARSE as the research platform, this dissertation studies performance optimizationissues through limiting the optimistic of event sending and reducing rollbackoverhead. Two methods of limiting optimistic are evaluated and compared, and then anew event-based rollback method is proposed based on the analysis of various existingrollback methods. Event-based rollback is the combination of several existing rollbackmethods. The idea is to use each rollback method in its best suitable place. Experimentshows that this new method achieved a better performance than any of the existingrollback methods used individually.Too much computing load is the main factor prevents the exploratory analysis(EA)method from being widely used. This dissertation adopts a cloning method to improve thecomputation effciency in EA. Simulation cloning improves the effciency through sharingthe common computation among multiple simulations. Incremental cloning mechanismis studied in detail and implemented in PARSE and the problems of cloning simulationon distributed-memory processors are solved. The performance of simulation cloning isalso evaluated, and result shows a remarkable performance improvement.At last, the methods of parallelizing sequential simulation model is discussed in order to simplify the development of parallel simulation model, and the parallelizationof Simulink model based on RTI and JMASE model based on PARSE are studied andimplemented.