Research On CFD Oriented Parallel Optimization Technology

CFD is an important content in the field of scientific computing,and has important applications in many fields.With the continuous development of computing technology,scientific computing has become the third major scientific research methods after theoretical research and scientific experiment.Scientific computing requires huge computational resources,in recent years the rapid development of high performance computers can make scientific computing using computational solution requires the use of scientific tests of the huge cost way to solve the problems of traditional and scientific computing applications in scientific research is more and more widely.As an important part of scientific computing,CFD plays a more and more important role in people's production and life.Based on the world's leading chip architecture,this paper presents a parallel optimization algorithm for the typical algorithms of CFD finite volume method.The solution to solve the problem of grid structure,has developed many solving methods,parallel optimization of LU-SGS finite volume algorithm in this paper is widely used in commercial software,analyzes the characteristics of LU-SGS algorithm method and feasible parallel optimization scheme is proposed for the LU-SGS algorithm characteristics of parallel optimization algorithm,the domestic nuclear structure line based on experiment show that the algorithm can line the 3D data related problems and optimization,improve the computational efficiency of the LU-SGS algorithm.For solving the problem of unstructured grid,the pretreatment time is long,especially the grid scale reached tens of millions or even over 1000 million pretreatment when mapping time problems,proposed the establishment of a global mesh adjacency fast search algorithm based on dynamic master-slave parallel graph reduction pretreatment process of storage and calculation time,the experimental results show that this method can significantly reduce the amount of complex grid problems of unstructured grid mapping pretreatment time,and has good parallel scalability.Combined with the analysis of the typical CFD algorithm,combined with the characteristics of CFD computing,we develop a parallel application platform for CFD applications,and provide users with visual job creation and management interface.The basic calculation process of CFD module package,the user can according to the application demand of customized workflow itself,and real-time monitoring and management of the calculation process,reduce the difficulty of the use of CFD application system,so that the industry users can put more energy into the industry application rather than the underlying implementation,improve the usability of CFD application system.At the same time,the platform i.s compatible with heterogeneous computing system,which has good portability and scalability,and provides users with a convenient and easy-to-use interface for high performance computing resources.