Jinv: A Parallel Method for Distributed Matrix Inversion

The past few years have seen the Grid maturing towards a final product, along with an increased popularity with researchers simulating their projects which contain some form of matrix inversion. The inversion of a matrix is very resource intensive and is often limited to either a Grid nodes CPU or memory capacity. This paper presents Jinv; a distributable matrix inversion solution. Developed using Jama, a linear algebra package for Java; it adopts a divide and conquer approach to tackle the main part of LU decomposition, matrix multiplication. Having implemented Jinv, initial tests have shown Jinv to be scalable and provide a working proof of concept. However it highlights the need for a more memory bandwidth efficient method to eliminate memory racing when executed on a single node.