MPI-SIM: using parallel simulation to evaluate MPI programs
Proceedings of the 30th conference on Winter simulation
Performance prediction of a parallel simulator
PADS '99 Proceedings of the thirteenth workshop on Parallel and distributed simulation
Dynamic component substitution in web-based simulation
Proceedings of the 32nd conference on Winter simulation
Optimizing Costs of Web-based Modeling and Simulation
IPDPS '01 Proceedings of the 15th International Parallel & Distributed Processing Symposium
Web-based simulation 2: performance prediction of dynamic component substitutions
Proceedings of the 34th conference on Winter simulation: exploring new frontiers
Performance prediction of large-scale parallel discrete event models of physical systems
WSC '05 Proceedings of the 37th conference on Winter simulation
Hi-index | 0.00 |
Multi-resolution models can be statically (i.e., before simulation) or dynamically (i.e., during simulation) abstracted to accelerate the simulations without compromising the analysis goals. However, abstractions must be carefully chosen because not all abstractions improve performance. Unfortunately, identifying performance enhancing transformations, particularly in parallel simulations, is a complex task. We have been investigating this bottleneck using a component-based, Time Warp synchronized modeling and simulation environment called WESE in which static and dynamic abstractions are performed using a methodology called Dynamic Component Substitution (DCS). Our ongoing research has resulted in further advances to a novel DCS Performance Prediction Methodology (DCSPPM). DCSPPM combines and compares platform-specific performance characteristics of components via static analysis of the model to predict performance changes due to DCS transformations. This paper presents the enhanced version of DCSPPM and experiments. Empirical evidence obtained using diverse models indicates that DCSPPM provides good estimates (error \le 5%) of the performance impacts of DCS transformations.