Benchmark Test of High Performance Computing Cluster Based on HPCC

Abstract

High performance computing has become one of the fundamental scientific and technological strengths, and it is also an indispensable tool for scientific and engineering computing. Benchmark test is very important for high performance computing. As a supplementary tool for HPL testing, High Performance Computing Challenge (HPCC) not only evaluates the floating-point computing power, but also comprehensively evaluates memory access speed, network communication speed and other performance. The HPCC testing will reflect the performance of all aspects of a high performance computing system. This paper takes a set of high performance computing clusters in the Gansu Computing Center to study the impact of the key parameters including matrix size N, the matrix block size NB and the two dimensional processor grid P×Q on the HPCC benchmark test, and conduct multi-node test and analysis of this cluster.