Skip to Main Content
MATLAB® and its open-source implementation Octave have proven to be one of the most productive environments for scientific computing in recent years. There have been multiple efforts to develop an efficient parallel implementation of MATLAB including by Mathworks® (Parallel Computing Toolbox), MIT Lincoln Labs (pMatlab) and several other organizations. However, most of these implementations seem to suffer from issues in performance or productivity or both. With the rapid scaling of high-end systems to hundreds of thousands of cores, and discussions of exascale systems in the near future, a scalable parallel Matlab would be of immense benefit to practitioners in the scientific computing industry. In this paper, we first describe our work to create an efficient pMatlab running on the IBM BlueGene/P architecture, and present our experiments with several important kernels used in scientific computing including from HPC Challenge Awards. We explain the bottlenecks with the current pMatlab implementation on BlueGene/P architecture, specially at high processor counts and then outline the steps required to develop a parallel MATLAB/Octave implementation, p2Matlab, which is truly scalable to hundreds of thousands of processors.
Date of Conference: 16-20 May 2011