高性能科学计算的特征分析及其实用方法研究

Study of Analysis of the Features Based on High Performance Science Computing and its Practical Methods

任何高性能科学计算(HPC)课题都是一项复杂的系统工程,其具体的应用效率要受到来自硬件和软件等许多因素,主要如并行算法、流水线技术、层次存储器技术和网络互联结构等的制约,诸因素既互相独立又互相关联。本文从一个典型的高性能科学计算——格点量子色动力学研究模型入手,在分析了HPC所涉硬软件存在的一些共性特征的基础上,总结出一些能够改善高性能科学计算应用效率的方法。通过对这些方法长期的应用实践和专门实验,证明它们是有效的。

Any one of HPC＇s projects is a complicated systematic engineering and its efficiency in practical application is restricted by many factors from hardwares and softwares such as parallel algorithm, pipelining technology, hierarchical memory technology and network connection structure which are both independent of and interrelated to each other. This paper,by means of studying the typical HPC--Lattice Quantum ChromoDynamic（LQCD） model and analyzing the common features of the hardwares and softwares involved in the HPC, works out some methods which can enhance the application efficiency of HPC. The long-period application and special experiments prove that these methods are effective.