A framework for parallel simulation application performance evaluation and optimization

Massively parallel simulation application software running on high performance computers is composed by a number of distributed sub-systems which have close interaction and logic processing relationship. How to analyze application performance problems during programming and execution, and how to effectively improve the performance of parallel simulation application are important and challenging problems. To meet the requirements of parallel simulation application software on system quality, system load balancing, system performance and other evaluation and optimization, the paper presents a framework for PSoA-Tool (Parallel Simulation-oriented Analyze Tool) on high-performance computer environments. It is divided into five layers: model and code input layer, data collection layer, data management layer, data analysis layer, and decision optimization layer. It can achieve performance optimization for massively parallel simulation application software in the code, node, and task levels by PSoA-Tool.