GAMMA: Global Arrays Meets MATLAB∗

MATLAB has become the dominant high-level language for technical computing. However, MATLAB has significant shortcomings when used for large-scale computationally intensive applications that require very high performance and/or significant amounts of memory. In such contexts, it is common for MATLAB to be first used for prototyping the application, followed by re-implementation in Fortran or C, using MPI for parallel execution. In this paper we describe GAMMA, a parallel global shared-address space framework for MATLAB, built on top of the Global Arrays library suite from Pacific Northwest National Laboratory. GAMMA enables the convenient development of large-scale computationally intensive applications directly in MATLAB, without the need to re-code them to achieve high performance. Preliminary experimental results on a Pentium cluster are provided, that demonstrate the effectiveness of the developed system.