Research On Parallel Performance Tuning Of CICE Based On "Sugon" And "Sunway Taihu Light" Supercomputer

Sea ice is one of the most sensitive parts of global climate change.Sea ice change has the role of global climate change indicator and amplifier.CICE as one of the most important components in the CESM has been widely used in computing in recent years.Climate and environmental issues in the Arctic and Antarctic are becoming increasingly complex.Nowadays,the climate and environmental problems of the North and South Arctic are becoming more and more complex,and the accuracy requirements in the calculation process of the CICE are becoming higher and higher.Since the application of the CICE itself is computationally intensive,the simulation calculation process of the CICE generally takes several hours to several days,and the timeliness of the calculation of the CICE is also increasing.How to improve the sea ice the computational efficiency of the model and exploring the best performance of the CICE have become hot topics in the industry."Sugon" supercomputer is a typical x86 architecture supercomputer,which realizes the operation of large-scale application programs through the multi-node cooperative working mode.This article transplants the 5.12 version of the CICE application to the "Sugon" supercomputer;the input data of the CICE open source is 3 degree grid data with lower accuracy,and the 3 degree grid data can only be divided in cartesian In the mode,in order to make the CICE simulation results more accurate,the T62 original grid data of the atmosphere is used to generate the 1 degree input data required by the CICE.Based on the seven grid division methods of the CICE,it is opened by changing the number of MPI processes,adjust the meshing method and related parameter configuration,and optimize the parallel performance of the CICE.As a new generation supercomputer with independent intellectual property rights,the" Sunway TaihuLight" supercomputer has a new architecture for the operation of the CICE.In order to enable the CICE to run successfully on the "Sunway TaihuLight" supercomputer and have better computing performance,the CICE was transplanted on the "Sunway TaihuLight" and parallel performance tuning.First,based on the "Sunway 26010" multi-core processor architecture and the characteristics of its unique compilation system,the CICE developed under the x86 architecture was successfully transplanted to "Sunway TaihuLight" by changing different operating modes Obtain the performance change of each operating mode with parameter configuration,and optimize the performance of the CICE so that has a good performance in"Sunway TaihuLight".Through performance tuning research,it was found that under the slenderx2 partitioning method,the performance of 320 MPI processes was improved by 8.32 times compared with 16 MPI processes.On the "Sunway TaihuLight" supercomputer,through performance tuning research,it was found that under 320 MPI processes,the performance of slenderx2(2x96 2)and slenderx2(2x6 32)increased by 117.3 times.It provides help for relevant researchers to use the CICE more efficiently,and also provides a reference for the parallel optimization work of the CICE.