Authors: Monte Lunacek and Thomas Hauser

Python is a popular interpreted programming language that is slowly emerging as an effective way to run distributed tasks in a High Performance Computing environment. This is partly because it’s much faster to write a parallel application in python than it is in a more traditional distributed language. Python also does not have the steep learning curve that makes using MPI with C or Fortran difficult for those unfamiliar with these more structured languages. As a consequence, users with some python modeling background can quickly bring their code to scale. In this talk, we will discuss some of the ways we have been using Python recently at the University of Colorado. Topics for the talk include interactive parallel development and education with the iPython notebook, task management with mpi4py and other distributed python implementations, and linking python models with C and Fortran to create efficient parallel applications. Our performance comparisons suggest that Python is an efficient way to glue a parallel application together.