Design And Application Of High Performance Computing Platform

With the rapid development of computer technology and network technology,High Performance Computing (HPC) has gradually become an important means oftechnological innovation and the third scientific method of research for peopleunderstanding and transforming the world. It has been widely applied on variousfields. However, there are many problems in the application of HPC onComputational Fluid Dynamics (CFD), especially for the large complex problems,leading topics and scientific issues when numerical simulation codes need to bewritten. This paper presents a parallel method based on Message Passing Interface(MPI). A set of HPC platform is designed and realized at Linux system, and appliedon one case.According to the characteristics of MPI, the domain decomposition method isapplied on the parallel numerical simulation. The data file of the whole region isdecomposed to a set of data files with small region, whose identification number has aone-to-one correspondence to the thread number in MPI. The small data files aredistinguished by adding an identification number in front of the filename. Theresearchers need to deal with the data in each small region and exchange the data onthe border, thus a simplification of design and program of the parallel code isachieved.Based on the requirement analysis of the HPC application, this paper designs aplatform, on which the total parallel computing operations can be done only in onenode. By using the Shell language and AWK language in Linux system, the operationsof file manipulation, parallel computation, process management, documentmanagement, user management and system management are realized through one keyoperation, thereby reducing the manual operations before and after the parallelcomputing.This platform is tested in a high performance parallel cluster with parallelenvironment configuration. The parallel program is designed and implemented byusing MPI and FORTRAN language, and the platform is applied on one case. Theresults show that, this platform successfully achieves a high performance parallelcomputing, reduces the difficulty of parallel computing, decreases the operations ofparallel computing, and has a high practical value.