Performance Evaluation of Parallel Data-Intensive Simulations Based on Graph Partitioning Approach

Scientific simulations require both increasing computing and storage power, parallel and distributed computing are strongly recommended to deal with them. Despite the power of the parallel approach to solving complex simulations, it is more difficult and error-prone, therefore, several factors create significant programming challenges and have the effect of reducing performance, which must be carefully managed to achieve high parallelism. This paper aims, on the one hand, to evaluate the performance of parallelism based on the graph partitioning approach, and on the other hand, to propose a performance analysis methodology that aims to help scientists to investigate the impact of inter-processor communication on the performance of an application First, a multilevel k way algorithms are used to compute a distribution of the calculations and associated data of simulation. Second, a performance evaluation strategy is used to investigate the efficiency of parallelism. Finally, we examine the results of the experiment and explain the perspective of this study.