高性能计算集群系统的设计和实现

主要论述了运用Oscar 2.3.1集群管理软件设计和搭建基于LinuxRedhat9.0的高性能计算集群系统的过程,并以支持集群并行计算的VASP包为例,通过对8个原子的立方结构C超原胞、64个原子的立方结构C超原胞和18个原子的六方结构ZnO超原胞的自洽计算,验证集群系统在计算性能上的优越性,同时提出了进一步提高集群性能的方案,以满足大型的科学计算和数值模拟.

Design and Realization of HPC Cluster

Using Oscar-2.3.1 and on the basis of Redhat 9.0 Linux,we design and realize a high performance computing (HPC) cluster system with one server and 8 clients.A VASP package supporting parallel computing in cluster system is used to calculate the self-consistent time of cubic carbon and hexagonal ZnO with different number of atoms in a super primitive cell.It is found that the performance of the cluster system using parallel computing increases with number of nodes,which is more distinct as the super primitive cells used become larger.This shows the advantage of HPC cluster in massive computation.To meet the challenge of massive computation and simulation in physical research,the power of HPC cluster can be further improved through constant updating both in software and hardware,which can be achieved more easily in comparison with that of supercomputers.One example is the optimization of parallel package and the introduction of a 1 000 M switcher and 1 000 M network cards to accelerate the communicating speed.