Matrix Multiplication Algorithm Cluster Environments Research And Design

In many fields of modern science, we often encounter problems of matrix computation.While the problem of matrix multiplication is the most basic of the kernel matrix calculation. As an extremely important mathematical solving tool, Matrix multiplication has been widely used in scientific research and engineering analysis, such as the solution of linear equations, engineering control, graphics and image processing etc. So the matrix multiplication time affects application time complexity computation. Increase the speed of matrix multiplication implementation for various industries have very important significance.In this paper, based on MPI parallel computing environment, in order to give full play to the effect of parallel computer, the purpose is improve the operation efficiency, through the analysis of factors influencing the matrix multiplication, it is concluded that when computing cluster to determine the cost of a parallel program, the good and bad of parallel algorithm selection, the program execution scheduling algorithm in the process of application, transfer protocol becomes the important factors that affect performance of matrix multiplication. This paper mainly studied the principle of matrix multiplication algorithm, the principle of load balancing algorithm, the principle of transmission protocol, several partition algorithm implementation are introduced in detail, the realization of load balancing strategy and algorithm as well as the advantages and disadvantages, and finally designed a study rotary weighting algorithm to achieve dynamic load balancing. And verify the effectiveness of the above algorithm.