Training Computational Scientists to Build and Package Open-Source Software

Abstract

High performance computing training and education typically emphasizes the first-principles of scientific programming, such as numerical algorithms and parallel programming techniques. However, many computational scientists need to know how to compile and link to applications built by others. Likewise, those who create the libraries and applications need to understand how to organize their code to make it as portable as possible and package it so that it is straightforward for others to use. These topics are not currently addressed by the current HPC education or training curriculum and users are typically left to develop their own approaches. This work will discuss observations made by the author over the last 20 years regarding the common problems encountered in the scientific community when developing their own codes and building codes written by other computational scientists. Recommendations will be provided for a training curriculum to address these shortcomings