The Research Of Computational Model Of Parallel Matrix Multiplication Based On DataMPI

Nowadays, the use of parallel computing become extensively. A reason for this is that the amount of data rising fast. Another reason for this is that the precision of the calculation results of people requirements increasing. The research on parallel computing model is one of the important research contents in the field of parallel computing, but until today, there is not a uniform, highly abstract parallel computing model. The complex architecture of parallel computing system is an important reason for this problem. However for a specific parallel application, we can study its calculation model according to its application characteristics, aim at improving the efficiency of this application.Matrix multiplication plays an important role in scientific computing. And its time complexity and space complexity are high. This paper study the application of matrix multiplication on MPICH and Data MPI. Data MPI combines the efficient communication mechanism of MPI and the high-performance data manage mechanism of Hadoop. We propose a parallel matrix multiplication programming model and program execution model based on Data MPI. According to the program execution model, the time function of matrix multiplication program is summed up.Specific work in this paper focuses on the following points:1. The parallel mechanism of Data MPI have been studied and analyzed. It mainly includes communication and parallel mechanism, data transfer mechanism, the source data input mechanism and so on. The influence factors of Data MPI on parallel matrix multiplication applications are summarized.2. Three parallel matrix multiplication algorithm are implemented and some parallel test are completed on the MPICH3. Then the method is givend which can evaluate the communication time and computing time at runtime.3. The programming model of parallel matrix multiplication based on Data MPI have been studied. We analyze the main idea of Data MPI implementation and its programming model and a parallel matrix programming model based on Data MPI is proposed. The aim is to simplify the implementing of matrix multiplication algorithm on Data MPI, and provide reference for other applications.4. In this paper M-FLOCA model is proposed. It is a parallel matrix multiplication program execution model based on Data MPI. The model consists of five levels, which are the file system layer F, the local storage layer L, the computing layer O, the communication layer C and the aggregation layer A. It constructs the execution process of parallel matrix multiplication on Data MPI.5. On the basis of the analysis of Data MPI parallel mechanism and its execution model, the calculation method of running time of the program is derived through the analysis of the time in the program execution process.