Implementation of dense matrix multiplication on 2D mesh

Abstract

A large number of algorithms have been developed for solving large dimension matrix multiplication through parallel computation. Lots of algorithms have been developed keeping performance matrices such as speed up, efficiency, isoefficiency etc. in linear order. We have compared the performance of simple block checkerboard partitioning algorithm with cannon's algorithm over 2D mesh topology in HPC Maverick (Rocks 5.4) by taking the mathematical problem matrix multiplication. Till the date not any of the algorithms clearly claimed to be superior then the others. It seems to be advantageous to partition matrix into blocks for multiplying on the 2D Mesh.