Asynchronous parallel simulation of parallel programs

摘要

Parallel simulation of parallel programs for large datasets hasnbeen shown to offer significant reduction in the execution time of manyndiscrete event models. The paper describes the design and implementationnof MPI-SIM, a library for the execution driven parallel simulation ofntask and data parallel programs. MPI-SIM can be used to predict thenperformance of existing programs written using MPI for message passing,nor written in UC, a data parallel language, compiled to use messagenpassing. The simulation models can be executed sequentially or innparallel. Parallel execution of the models are synchronized using a setnof asynchronous conservative protocols. The paper demonstrates hownprotocol performance is improved by the use of application-level,nruntime analysis. The analysis targets the communication patterns of thenapplication. We show the application-level analysis for message passingnand data parallel languages. We present the validation and performancenresults for the simulator for a set of applications that include the NASnParallel Benchmark suite. The application-level optimization describednin the paper yielded significant performance improvements in thensimulation of parallel programs, and in some cases completely eliminatednthe synchronizations in the parallel execution of the simulation model