Supercomputing with MPI meets the Common Workflow Language standards: an experience report

TL;DR

This paper reports on integrating MPI parallel applications with the Common Workflow Language standards, highlighting implementation details, benefits, and community feedback for future standardization.

Contribution

It introduces a simple extension to CWL for MPI applications, demonstrating practical integration and benefits within high-performance computing workflows.

Findings

Enhanced compatibility with HPC performance tools

Improved interfacing with HPC module systems

Successful implementation of MPI workflows in CWL

Abstract

Use of standards-based workflows is still somewhat unusual by high-performance computing users. In this paper we describe the experience of using the Common Workflow Language (CWL) standards to describe the execution, in parallel, of MPI-parallelised applications. In particular, we motivate and describe the simple extension to the specification which was required, as well as our implementation of this within the CWL reference runner. We discuss some of the unexpected benefits, such as simple use of HPC-oriented performance measurement tools, and CWL software requirements interfacing with HPC module systems. We close with a request for comment from the community on how these features could be adopted within versions of the CWL standards.