DCU cluster-oriented large-scale finite element grid parallel partitioning method and device

Abstract

The invention relates to the technical field of high-performance computing, in particular to a DCU cluster-oriented large-scale finite element grid parallel partitioning method and device. The method comprises the steps: discretizing a solving area into grids, and describing the distribution situation of the grids obtained through discretization through a weighted dual graph, wherein vertexes in the weighted dual graph represent the grids, edges in the weighted dual graph represent the connection relation between grids, and the dividing speed of the recursive spectral dichotomy can be greatly improved through parallelism of sub-graph dividing tasks; carrying out parallel calculation of hot spots of the spectral dichotomy on the basis of parallelism of sub-graph dividing, so the processing time of finite element grid dividing is further shortened, the problem that the time demand for dividing the large-scale grids is too long in a serial recursive spectrum dichotomy method is solved, and the efficiency of solving the problem is greatly improved.

Description

technical field [0001] The invention relates to the technical field, in particular to a method and device for parallel partitioning of large-scale finite element grids oriented to DCU clusters. Background technique [0002] The principle and basic concept of the finite element method is to replace complex problems with simpler problems and solve them. It decomposes the solution domain into many small interconnected subdomains called finite element grids, and assumes a suitable approximation for each unit. solution, and then obtain the final solution of the problem according to the conditions that the solution domain as a whole needs to satisfy. The discretization of the solution domain makes the finite element method not only have better calculation accuracy, but also adapt to various complex shapes. Relying on the massive computing throughput capacity of supercomputers and large-scale parallel computing technology, the finite element method has become an effective way to s...

AI Technical Summary

Problems solved by technology

Among various area division algorithms, the recursive spectral dichotomy method has the characteristics of load balance and good topology compared with traditional algorithms and other recursive dichotomy methods, and is very suitable for dealing with high-precision simulations. is very long and grows substantially with the number of meshes

[0004] Taking the serial grid partition module established based on the recursive spectral dichotomy as an example, the test results can be as follows figure 1 As shown, it takes nearly 11 hours to process a 10 million-scale grid partition, and it may even take several days for a 100 million-level grid. It can be seen that the serial recursive spectral dichotomy method has the potential to divide large-scale grids. The time requirement is too long, resulting in greatly reduced efficiency in solving problems

Images