Analysis of MPICH Communication Interface and Optimization of Data Partitioning in Matrix Calculation

This paper presents optimal data partitioning strategy to improve the efficiency in parallel matrix calculation. We analyze communication delay in communication layer, structure of communication library in software layer. And three parallel algorithms of matrix calculation are proposed. Theoretical analysis and empirical results show that matrix partitioning strategy are computionally more efficient than traditional.

[1]  Robert B. Ross,et al.  Using MPI in high-performance computing services , 2013, EuroMPI.

[2]  Yannick Saouter.  Heuristics for cannon's algorithm with an application to lyons sporadic group , 2011, ACCA.

[3]  Yunchun Li,et al.  Optimization of Parallel I/O for Cannon's Algorithm Based on Lustre , 2012, 2012 11th International Symposium on Distributed Computing and Applications to Business, Engineering & Science.

[4]  Jens Meiler,et al.  Comparative analysis of machine learning techniques for the prediction of logP , 2011, 2011 IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB).

[5]  Chee-Kong Chui,et al.  The LogP and MLogP models for parallel image processing with multi-core microprocessor , 2010, SoICT.

[6]  Pavan Balaji,et al.  Implementing the MPI-3.0 Fortran 2008 Binding , 2014, EuroMPI/ASIA.

[7]  Mark A. Taylor,et al.  Architecture of LA-MPI, a network-fault-tolerant MPI , 2004, 18th International Parallel and Distributed Processing Symposium, 2004. Proceedings..