GRAPES system optimization method based on parallel supercomputing grid cloud platform

Abstract

The invention relates to a GRAPES system optimization method based on a parallel supercomputing grid cloud platform. The method comprises the steps that (S1) a test dataset is loaded, a system is run,and (S1.1) a system-level test, (S1.2) a communication-level test and (S1.3) a function-level test are performed separately, wherein in the (S1.3) function-level test, a called function is monitored,and running features of the function are acquired; (S2), test result analysis is performed according to an exported system feature file, wherein test result analysis comprises (S2.1) system test result analysis, (S2.2) MPI communication-level test result analysis and (S2.3) function-level test result analysis; and (S3) optimization processing is performed according to analysis results, wherein optimization processing comprises vectorization, load balancing and replacement of a function in a GRAPES_GFS by use of a library function. Through the method, optimization of Grapes on the parallel supercomputing grid platform is realized, and system running efficiency is improved.

Description

technical field [0001] The invention relates to the field of meteorological data processing, in particular to a GRAPES system optimization method based on a parallel supercomputing grid cloud platform. Background technique [0002] After long-term development, the scientific basis and technical methods of numerical weather prediction have been relatively mature, and it is the most important scientific way to make weather forecasts, and has unique advantages in the forecasting of extreme weather events. In addition to fully considering the contribution of atmospheric motion at various scales, climate numerical forecasting must also take into account the interaction between the atmosphere and other spheres of the earth. Therefore, the amount of data involved in the calculation is very large, and the timeliness of credible medium-term numerical weather is further extended. , based on the higher and higher requirements for real-time performance of numerical forecasting, it is pa...

AI Technical Summary

Problems solved by technology

[0005] However, with the continuous improvement of the numerical weather prediction model, the continuous improvement of the resolution, and the gradual development of the forecast towards the extended-range forecast and short-term forecast, the GRAPES system runs on the Sunway 4000A platform, and the computing power and resolution of the platform are limited. The demand is also constantly expanding. Therefore, when running the GRAPES system in the existing operating environment, there will be problems of low calculation efficiency such as slow calculation speed and long calculation cycle due to the sharp increase in calculation amount and limited calculation capacity; on the other hand , the GRAPES mode is a grid mode, which adopts the general horizontal grid division in the parallel decomposition of the calculation area, and in order to improve the accuracy of the forecast, the calculation grid will be divided smaller and smaller, and the corresponding calculation amount will be increase

Similarly, because the computing power of the Sunway 4000A platform is

Images