Data Aware Simulation of Complex Systems on GPUs

Abstract

GPUs have been demonstrated to be highly effective at improving the performance of Multi-Agent Systems (MAS). One of the major limitations of further performance improvements is in the memory bandwidth required to move agent data through the GPU’s memory hierarchy. This paper presents a formal model for data aware simulation and an empirical study into the impact of minimising data movement on performance. This study proposes a method that can be applied to the simulation of complex systems on GPUs to extract required data from agent behaviour during simulation time and how this information can be used to reduce data movement. The FLAME GPU software has been extended to demonstrate this technique. Three benchmark experiments have been applied to evaluate the overall reduction in simulation execution time under specific criteria. The results of the comparison between the current and new system show that reducing data movement within a simulation improves overall performance with up to 4.8x speedup reported.