基于Linux的高性能集群的构建和性能优化

Construction and optimization of a Linux-based high performance computing cluster

提出并实现了一种并行计算集群搭建的方案,采用国际通用的HPL标准进行浮点运算性能的测试。在优化集群高性能运算性能的方面,对传统基于经验的穷举办法进行优化,提出一种基于参数影响因子排序的优化方案,有效地减少了穷举法盲目试验的时间。另外次算法还具有一定的开放性,以往的经验数据可以得到有效利用。

This paper proposed and implemented a scheme for high performance computing cluster,then measured the floating-point performance of the cluster with the internationally accepted standard benchmark HPL.To optimize the performance,an optimal approach based on sequencing of impact factor is proposed,which effectively reduced the cost of optimization comparing with the traditional experience-based exhaustion approach.Furthermore,this algorithm is open to some extent,as former empirical date can be utilized effectively.