Details zur Publikation |
| Kategorie | Textpublikation |
| Referenztyp | Qualifizierungsarbeiten |
| URL | https://nbn-resolving.org/urn:nbn:de:bsz:14-qucosa2-1046483 |
Lizenz  |
|
| Titel (primär) | Portable and reproducible container-based HPC-enabled simulations for coupled processes |
| Autor | Bilke, L.
|
| Erscheinungsjahr | 2026 |
| Department | ENVINF |
| Seite bis | xii, 98 |
| Sprache | englisch |
| Topic | T8 Georesources |
| Keywords | reproducibility, computational science, scientific computing, scientific workflows, software engineering |
| Abstract | Configuring and deploying complex scientific simulation software on local workstations or high-performance computing (HPC) environments remains a challenging task due to intricate interdependencies among compilers, libraries, and tools. Even after successful installation, reproducing simulations on other systems often proves difficult, as every software dependency must be recreated precisely, an endeavour that can be rendered impossible by differences in operating systems or evolving hardware architectures. These obstacles threaten the long-term verifiability and transparency of scientific results, especially in computational fields where simulation outcomes inform decisions of societal relevance. To overcome these challenges and enable reproducible research in scientific computing, this dissertation investigates Linux container technologies as a lightweight form of computer virtualisation and encapsulation. Recent advances in scalable, MPI-enabled container platforms such as Singularity (now Apptainer) allow for portable, container-based workflows on arbitrary HPC systems without loss of performance. By encapsulating all dependencies and runtime binaries into self-contained execution environments, containers form a cornerstone for ensuring reproducibility, portability, and longevity of scientific simulations. We demonstrate how transparent development practices, automated testing, and continuous integration establish the cultural and technical foundations for reproducible computational research on the example of the OpenGeoSys project. By utilising GNU Guix for software packaging we enable fully declarative, bit-for-bit reproducible software environments to be deployed as portable Apptainer container images and demonstrate result reproducibility on the bit-level of complex scientific workflows. Beyond the technical advances, this work responds to the growing demands of funding agencies and regulatory bodies for reproducible, transparent, and archivable scientific results, particularly in safety-critical domains such as nuclear waste management. By introducing frontier computer science methodologies into environmental sciences, this thesis contributes to the advancement of both disciplines, providing a blueprint for reproducible scientific workflows. |
| Bilke, L. (2026): Portable and reproducible container-based HPC-enabled simulations for coupled processes Dissertation TU Dresden, Dresden, xii, 98 pp. |
|