Bridging HPC Communities through the Julia Programming Language

This repository contains information and code to reproduce the results presented in the article bibtex @online{churavy2022bridging, title={Bridging {HPC} Communities through the {J}ulia Programming Language}, author={Churavy, Valentin and Godoy, William F and Bauer, Carsten and Ranocha, Hendrik and Schlottke-Lakemper, Michael and R{\"a}ss, Ludovic and Blaschke, Johannes and Giordano, Mos{\'e} and Schnetter, Erik and Omlin, Samuel and Vetter, Jeffrey S and Edelman, Alan}, year={2022}, eprint={2211.02740}, eprinttype={arxiv}, eprintclass={cs.DC}, doi={10.48550/arXiv.2211.02740} }

If you find these results useful, please cite the article mentioned above. If you use the implementations provided here, please also cite this repository as bibtex @misc{churavy2022bridgingRepro, title={Reproducibility repository for "{B}ridging {HPC} Communities through the {J}ulia Programming Language"}, author={Churavy, Valentin and Godoy, William F and Bauer, Carsten and Ranocha, Hendrik and Schlottke-Lakemper, Michael and R{\"a}ss, Ludovic and Blaschke, Johannes and Giordano, Mos{\'e} and Schnetter, Erik and Omlin, Samuel and Vetter, Jeffrey S and Edelman, Alan}, year={2022}, howpublished={\url{https://github.com/JuliaParallel/paper-2022-HPC}}, doi={10.5281/zenodo.7236017} }

Abstract

The Julia programming language has evolved into a modern alternative to fill existing gaps in the requirements of scientific computing and data science applications. Julia comes with a unified and coordinated single-language and ecosystem paradigm, and has a proven track record at achieving high-performance without sacrificing user productivity. This makes it a viable alternative to the existing increasingly costly many-body workflow composition strategy in high-performance computing (HPC), i.e., using traditional HPC languages (Fortran, C, C++) for simulations and higher-level languages (Python, R, Matlab) suitable for data analysis. Julia’s rapid growth in language capabilities, package ecosystem, and community make it a promising new universal language for HPC. This paper presents the views of a multidisciplinary group of researchers in academia, government, and industry advocating for an HPC software development paradigm that emphasizes developer productivity, workflow portability, and low barriers for entry. We believe that the Julia programming language, its ecosystem, and its community, provide modern and powerful capabilities which enable this groups objectives. Crucially, we believe that Julia can provide a more feasible and less costly approach to program scientific applications and workflows targeting HPC facilities. We examine the current practice and role of Julia as a common end-to-end programming model to address major challenges in scientific reproducibility, data-driven artificial intelligence/machine learning (AI/ML), co-design and workflows, scalability and performance portability in heterogeneous computing, network, data management, and community education. As a result, we consider necessary the diversification of current investments to fulfill the needs of the upcoming decade as more supercomputing centers prepare for the Exascale era.

Numerical experiments

To reproduce the numerical experiments, you need to install Julia. The subfolders of this repository contain README.md files with instructions to reproduce the numerical experiments, including postprocessing.

Authors

• Valentin Churavy, Massachusetts Institute of Technology (MIT), US
• William F Godoy, Oak Ridge National Laboratory (ORNL), US
• Carsten Bauer, Paderborn Center for Parallel Computing (PC2), Germany
• Hendrik Ranocha, University of Hamburg, Germany
• Michael Schlottke-Lakemper, RWTH Aachen University, Germany
• Ludovic Räss, ETH Zürich, Switzerland
• Johannes Blaschke, National Energy Research Scientific Computing Center (NERSC), US
• Mosé Giordano, University College London (UCL), UK
• Erik Schnetter, Perimeter Institute, Canada
• Samuel Omlin, Swiss National Supercomputing Centre (CSCS), Switzerland
• Jeffrey S Vetter, ORNL, US
• Alan Edelman, MIT, US

Disclaimer

Everything is provided as is and without warranty. Use at your own risk!