Accelerating Large-scale DEVS-based Simulation on the Cell Processor

摘要

This paper presents a new technique for efficient parallel simulation of large-scale DEVS-based models on the IBM Cell processor, which has one Power Processing Element (PPE) and eight Synergistic Processing Elements (SPEs). By taking a performance-centered approach, the technique allows for exploitation of multi-dimensional parallelism to overcome the bottlenecks in the simulation process. We illustrate the underlying design methodology with detailed simulation profiles. Our preliminary experiments have already produced promising results, accelerating the baseline PPE-only simulation of a fire model and a flood model by a factor of up to 70.6 and 83.32 respectively. The technique not only enables DEVS users to harness the potential of the Cell processor without being distracted by the technical complexity of multicore programming, but also provides insights on migration of legacy software to current and future multicore platforms.