Structural-dynamic-analysis explicit-different-step-length parallel computing method

Abstract

The invention provides a structural-dynamic-analysis explicit-different-step-length parallel computing method. The method includes the four steps that (1) data files required by parallel computing aregenerated with the partition multiple-node strategy; (2) time step lengths of all partitions are selected according to partition unit characteristics to be subjected to explicit prediction step computing; (3) in small-step-length partitions, partition-inside and boundary-node data is solved through step-by-step decreasing of the explicit sub-cycle process, and in large-step-length partitions, inside and boundary-node data is solved through the main time step length; (4) the boundary-node data required by paralleling is received by all the partitions, partition information is updated, strainsand stress are computed as required, and partition node data is output; if a main time step is ended, the program is stopped, and if a main time step is not ended, the step (2) is newly carried out. According to the structural-dynamic-analysis explicit-different-step-length parallel computing method, in the large-scale dynamic parallel analysis process, different time step lengths can be selectedaccording to the partition unit characteristics, partition different-step computing coupling is achieved with the multiple-node strategy, and parallel efficiency is effectively improved.

Description

technical field [0001] The invention relates to a design method in the field of supercomputer application technology, in particular to an explicit asynchronous long parallel calculation method for structural dynamics analysis. Background technique [0002] Theory, experiment and calculation are the three carriages that promote the development of science and technology. In recent years, with my country's continuous investment in research and development of supercomputers and technological development, a number of supercomputing systems represented by "Tianhe-2" and "Sunway TaihuLight" have sprung up, occupying the top 500 supercomputing systems in the world. Computing system tops the list. With the rapid development of hardware technology and numerical technology, computing technology has become a veritable engineering weapon, playing a vital role in the fields of aerospace equipment, deep-sea technology, nanomaterials, and biomedicine. [0003] Contrary to the gratifying br...

AI Technical Summary

Problems solved by technology

However, when this method is used for large-scale parallel structure analysis, the parallel efficiency of the system is not high.

This is because for large-scale problems, firstly, with the increase of the number of sub-regions, the scale and condition number of the interface equations also increase sharply, which leads to the need for more iterations to converge when solving, thus greatly increasing the total number of the system. Second, in this method, all partitions use the same size of time step, while the flexible multi-body system needs to finely divide the grid in some local areas, which will lead to the use of a smaller time step as a whole, which greatly increases the computational cost. In addition, this method does not consider the characteristics of the multi-core distributed environment. When it solves the interface equation, all sub-regional processes must participate in both local communication and global communication.

Due to the difference in communication delay between nodes and nodes, the inter-process communication and synchronization overhead will increase significantly with the increase of sub-regions.

Images