Large-scale CFD parallel computing method based on distributed Mysql cluster storage

Abstract

The invention provides a large-scale CFD parallel computing method based on distributed Mysqul cluster storage, and relates to communication between cooperating processes and data storage on multiple computing nodes. The large-scale CFD parallel computing method comprises the steps that the CFD process on each computing node be connected to a database through the fact that an API is driven by a Mysql, and initial data are acquired from a database to be used for computing; according to the CFD process, MPI protocol communication is used in the computing process, and the data or global reduction information on a zone boundary are / is exchanged; for data on each iterative step, the process is connected with a handle through the Mysqul to write the data into the database; after CFD parallel program computing is completed, Mysqul clusters are synchronized to obtain complete data in the computing process. Compared with an existing CFD parallel computing method, an open source Mysql database is fully used for achieving high performance and high reliability of data storage, and various information and data in the computing process can be flexibly and completely stored by CFD parallel programs. According to the large-scale CFD parallel computing method, a development process is simplified, the wall time of execution of the CFD parallel programs is reduced, the practicability is high, and popularization is easy.

Description

technical field [0001] The invention relates to a large-scale CFD program parallel realization method, especially a data storage method in the CFD parallel program. Background technique [0002] In the field of fluid simulation and calculation, with the increase of problem scale and the improvement of data accuracy requirements, the number of grids for 3D CFD simulation has rapidly increased from tens of millions to billions or even hundreds of billions. The resulting huge amount of calculation requires high-performance computing to be able to give the calculation results of the program within an acceptable time range. At present, parallel computing is the key and hot spot of high-performance computing, and MPI is a protocol for communication between parallel processes. MPI technology is a standard interface for cross-host process communication in distributed clusters, and has received strong support from various computer manufacturers and software developers. MPI-based I / ...

AI Technical Summary

Problems solved by technology

The intermediate data generated during the process calculation, the usual CFD parallel program will be directly discarded due to the cumbersome recording

If the file I / O method is used to record through the MPI interface, it cannot be recorded and stored flexibly because of its strict format requirements, which will cause a large amount of information to be discarded during the running of the process, which is not conducive to tracking the running status of the program Debugging and Profiling

Images