Enhancing Scientific Reproducibility: A Continuous Integration Workflow for High-Performance Computing

Abstract

Scientific software plays an increasingly important role in modern re- search. Yet, the lack of software development training among researchers com- bined with limited funding and recognition for software development often results in error-prone and difficult-to-maintain code. Moreover, the often performance- critical nature of scientific software adds additional complexity to the development and testing process as researchers need to work with high-performance computing (HPC) systems. In this paper, we address the issue of integrating and reproducing workflows on an HPC system by introducing a continuous integration (CI) workflow tailored for HPC environments. Our workflow combines three tools to simplify the execution and validation of computational tasks on HPC systems: Singulariy, HPC-Rocket, and Fieldcompare. We leverage Singulariy containers for consistency across computing setups, then utilize HPC-Rocket to launch and mon- itor a simulation on an HPC system from within a CI pipeline, and finally validate the results using Fieldcompare to ensure reproducibility.