Manufacturing Simulation Using Bsp Time Warp with Variable Numbers of Processors

The performance of an optimistic parallel simulation run depends on many factors. These factors range from the inherent complex nature of the simulation workload to the overhead in using the available parallel computing resources. Very often, using the maximum number of available processors for a parallel simulation run will not yield the best performance achievable. In this paper, we describe a new approach for executing BSP Time Warp optimistic parallel simulations using variable number of processors. Processors are automatically added or removed during the simulation runtime based on a performance cost model. Experiment results using this new approach on a real-world semi-conductor factory model is presented.