mpiPython: Prospects for Node Performance

Abstract

Python as an interpreted language is limited in performance by its ability to optimize code. With it being a high-level programming language, it’s still a strong choice for data scientists to learn and use. If Python could be optimized for parallel programming, its full potential in parallel and cloud computing environments could be achieved. mpiPython is a message-passing module that gives Python the ability to be used in SPMD (Single Program Multiple Data) environments. In this paper, we review basic features of mpiPython, including its runtime communication libraries and design strategies. mpiPython also has new features to help mpiPython programmers, that includes simplifying traditional MPI initialization. During the development of mpiPython, we realized that individual node performance of mpiPython is uncertain and critical. mpiPython node performance will be analyzed within the benchmarks.