Method, device and equipment for determining MPI parameters of parallel computing software

Abstract

The invention discloses a method, device and equipment for determining MPI parameters of parallel computing software. The method includes: obtaining an MPI platform set corresponding to the parallel computing software; running the parallel computing software in the test MPI platform according to a preset operation mode to perform parallel computing , and detect the operating characteristics of the parallel computing software; according to the operating characteristics, determine the software type of the parallel computing software; from the MPI parameter set corresponding to the software type stored in the database, determine the corresponding target MPI parameters of the parallel computing software; the present invention uses the operating characteristics Similar parallel computing software is divided into the same software type, and the target MPI parameters corresponding to the parallel computing software are determined from the MPI parameter set corresponding to the software type of the parallel computing software in the database, so as to realize the rapid selection of the MPI parameters of the parallel computing software, and the parallel computing software The parallel computing performance of the computing software performs fast optimization space prediction and optimization, shortening the tuning time of the parallel computing software.

Description

technical field [0001] The present invention relates to the technical field of computers, in particular to a method, device and device for determining MPI parameters of parallel computing software. Background technique [0002] With the development of modern social science and technology, the use of parallel computing software is becoming more and more extensive, such as CAE (Computer Aided Engineering) software in the manufacturing field and other parallel methods in the field of climate prediction and biochemical computing. software, etc. MPI (Massage Passing Interface) is a standard protocol for information exchange between computers during parallel computing across nodes. Parallel computing software (such as CAE software) realizes multi-node parallel computing through this protocol. MPI has different control parameters. Due to the different algorithms of various parallel computing software, each control parameter has different effects on the parallel computing performan...

AI Technical Summary

Problems solved by technology

At present, the comprehensive test method (that is, to test each parameter separately) is generally used for testing, which requires a lot of computing resources and consumes a lot of time and material resources.

Using the principle of orthogonal experiment can effectively reduce the number of tests. Through the orthogonal experiment design, the option parameter group selected in the orthogonal design table can be tested, and the optimization result can be obtained with a small amount of testing; but even so, the test workload is quite large. Considerable, coupled with the large number of parallel computing software, if a similar test is carried out for each parallel computing software, the workload is unacceptable

Images