POSE: Getting Over Grainsize in Parallel Discrete Event Simulation
International Conference on Parallel Processing (ICPP) 2004
Publication Type: Paper
Repository URL:
Abstract
Parallel discrete event simulation (PDES) applications encompass a broad range of analytical simulations. Their utility lies in their ability to model the system under study and provide information about the behavior of that system in a timely manner. Current methods for PDES provide limited performance improvements over sequential simulation. Many logical models for applications have highly irregular structures and fine granularity, making them challenging problems to parallelize. In the development of our Parallel Object-oriented Simulation Environment (POSE), we examined the overhead required for optimistically synchronizing events to obtain correct results in parallel. We then designed a new object model and an adaptive optimistic synchronization method to improve the performance of fine-grained PDES applications. These novel approaches exploit the speculative nature of optimistic protocols to achieve better single processor performance than a basic sequential approach, and show how scalability can be achieved for previously hard-to-parallelize fine-grained simulations.
TextRef
Terry L. Wilmarth and L. V. Kale, "POSE: Getting Over Grainsize in Parallel Discrete Event Simulation", To appear in 2004 International Conference on Parallel Processing, August 2004.
People
Research Areas