How parallelization helps crowd simulation: Study of an OpenMP-Based system

Abstract

This paper analyzes the parallelization efficiency of Menge [1], an open source virtual crowd simulation system widely used for algorithm benchmarking, with focuses on three aspects: performance of the existing parallel processing scheme, bottleneck of parallel processing, and improvement opportunities for parallel efficiency of the system. First, we calculate the speedup ratio of each Menge module by analyzing the data collected under with and without OpenMP scenarios. We then identify the bottleneck of the parallel computing from the empirical study. Secondly, the possibility of improving the performance through hardware configuration is analyzed by testing the performance of the system on different computers which have the similar clock frequencies but different number of cores. The experimental results show that there is still plenty of room for improvement in the parallelization performance of the system.