Network Parallel Computation Based on the Large Increment Method
|Keywords||King incremental step algorithm Generalized inverse matrix Parallel computing Cluster System MPI|
The first part of a more comprehensive and complete work on the system of the extraordinary incremental step algorithm. The main results are: (1) starting from the mechanical concepts elaborated the ideas of King incremental step algorithm derived from the classical force derivation given extraordinary incremental step algorithm format, including the problem of nonlinear elastic, elastoplastic general expression of formulas and finite element format; (2) compared LATIN algorithm has a certain similarity with King incremental step algorithm, pointed out that the force method and the generalized inverse matrix theory-based the LIM algorithm and LATIN algorithm grouping in the equation and iterative formal similarities, but have mechanical concepts and mathematical theory based on fundamentally different; King incremental step algorithm complete mathematical proof is given in (3), including the King incremental step algorithm for solving only proof of the stability of the convergence of the solution reconciliation, thus proving that the King incremental step algorithm in mathematical completeness; (4) for the convergence of the solution discussed algorithm selected sample points in elastoplastic problems, pointing out that the LIM algorithms guarantee convergence problems premise incremental method has greater freedom than in the choice of sample points. The second part of the thesis is also this thesis, but the main work is an the extraordinary incremental step algorithm parallelization research. The main content and results: (1) in-depth analysis of the the extraordinary incremental step algorithm parallel, pointed out the inherent parallelism of the parallel algorithm with the level of mechanical concepts and space on parallel algorithms in numerical level parallelism in the matrix-vector operations; (2) design asynchronous parallel algorithm of distributed storage model, the extraordinary incremental step algorithm, LIM algorithm master-slave style parallel program structure; (3) for the plane stress problem cases, design, and gives King incremental step algorithm for parallel programs based broadcast and non-blocking communication-based parallel programs; (4) describe the parallel environment used in the present paper, the hardware environment for the cluster system software environment for Linux operating system coupled with the MPI programming environment based on message passing; (5) King incremental step algorithm parallel programs two programs parallel performance tests, analyze the problems in the program, and then gives the procedures for communication optimization program cluster systems in the the MPI environment under the High Performance Computing Research Institute of Tsinghua University, Department of Computer Science, the results are given close to a speedup of 10 and more than 95% of node resources occupancy rate, the scale of the solution can reach thousands of unit grid, exhibit good parallel Efficiency.