Optimization of GRAPES System Based on Parallel Supercomputing Grid Cloud Platform

GRAPES is a multiscale universal system developed by China for global medium range numerical prediction. With the development of numerical weather prediction theory, the numerical weather prediction model is improved and the resolution is also increased. Forecasts are also being extended to extended and short-term forecasts. These reasons lead to the increase of computational data and the improvement of prediction accuracy, which means that the computing power and resolution requirements of the operating platform are also increasing. Coupled with four-dimensional variational assimilation, the pressure of the original operating platform is becoming larger and larger, and the original operating environment becomes the bottleneck. To solve this problem, the work about this paper is to transplant GRAPES system to a new high performance computing platform. This platform is called Milkyway-2, taking the INTEL X86 architecture as the core. In this paper, we first test the GRAPES-GFS model on a large scale. According to the test result, we then analyze the existing problems of this model. Finally, we formulate the corresponding optimization strategy. From the optimization results, the operating speed of the system has been improved greatly.