An approach to enhance the performance of large‐scalernstructural analysis on CPU‐MIC heterogeneous clusters

摘要

Clusters with the CPU‐MIC heterogeneous architecture are becoming more popular in recentrnyears. However, it is not easy to achieve good performance on such machines. The key challengernhas been the asymmetry within clusters, arising from different kinds of execution units as well asrndifferent communication latencies. To improve the performance of large‐scale structural analysisrnon CPU‐MIC heterogeneous clusters, a multi‐layer and multi‐grain collaborative parallel computingrnapproach is proposed in the paper. The proposed method combines the parallel algorithm andrnthe hardware architecture of CPU‐MIC heterogeneous clusters together. Through mappingrncomputing tasks to various hardware layers, it both resolves the load balance problem betweenrnCPU and MIC devices and significantly reduces the communication overheads of the system.rnNumerical experiments conducted on Tianhe‐2 supercomputer show that the proposed methodrnobtained better performance compared with the traditional approach. Scalability investigationrnshowed that the proposed method had good scalability with respect to problem sizes. The findingsrnof this paper are of help to the parallel porting and performance optimization of other applicationsrnon CPU‐MIC heterogeneous clusters.