Gast ::
| Hauptseite > Publikationsdatenbank > Reproducible HPC workflows using JUBE |
| Hauptseite > Publikationsdatenbank > Reproducible HPC workflows using JUBE |
| Conference Presentation (After Call) | FZJ-2026-02027 |
;
2026
Abstract: When executing workflows in computer-based research, often several steps need to be taken to obain a result, from downloading and compiling a software package, preparing input, creating job scripts to submitting those to an HPC system or executing the program on the local computer, and parsing respective output.Oftentimes, these workflows only differ in minimal ways and scripts for batch jobs on HPC systems are manually copied and adapted.Initiated as a software infrastructure for benchmarking high-performance computing (HPC) systems, the Jülich Benchmarking Environment (JUBE) has evolved to a flexible framework for creating workflows especially on HPC systems that enable users to automate workflows and measure and extract information during the workflow that help making these workflows more reproducible and standardized.While workflows in JUBE ensure repeatability, it is the responsibility of the user to specify all parameters needed to ensure the generation and extraction of relevant information about its workflows that help making them more reproducible. Once defined, the generation of such meta data is done automatically as part of the workflow.This half-day hands-on tutorial targets current or prospect users of HPC systems who want to automate workflows from building software, preparing inputs, generating and submitting job scripts, and analyzing output in a structured manner.The material is using the Carpentries Workbench format and is freely available online.Following the Carpentries-style of teaching, the participants will follow and code along an example workflow to automatically download, build, and execute the GROMACS molecular simulation code with an example input.The application itself only acts as a proxy example for a workflow involving several stepsAfter the course the participants will have experienced the generation of a complete workflow and will be able to write basic workflow patterns for their own HPC workloads.
|
The record appears in these collections: |
