Reusable Execution Environments in NFDIxCS

ReusableExecutionEnvironmentsin Fabian Huch1, Christoph Blessing2, Sabih Ahmed Khan2, Reza Salkhordeh3, Jan Bernoth4, Michael Goedicke5 1Technische Universtität München, 2Gesellschaft für wissenschaftliche Datenverarbeitung mbH Göttingen, 3Universität Potsdam, 4Johannes Gutenberg-Universität Mainz, 5Universität Duisburg-Essen Problem Modern science depends on software. Yet, reproducibility of research that involves software suffers from two main problems: •Source code and data of published results are frequently missing. •Setting up the required infrastructure can be very complicated and timeconsuming. This is a major problem, which recently led to the reproducibility crisis in the area of machine learing [2, 3]. Solution Reusable Execution Environments (REEs) allowresearcherstorunsoftwareinconditions identical to the original system, ensuring that results can be reliably reproduced. Challenges 1. To re-create the same environment, REEs must embed all bootstrap software or rely on long-term archives. 2. REEs specifications should be precise,butresearchersmustnotbeburdenedwithoverly complexlanguages or standards. 3. Run-time abstractions are required so REEs can be executed on future systems, virtualizing older architectures if necessary. 4. Software may depend on hardware specifics so re-creating an identical environment might not be feasible. Resources REE Web Service REEuse Tool Kinds of REEs Source-Defined Systems Containers Virtual Machines Software Bill of Materials •Software environments, built from source •Only sources need to be stored, e.g. in the Software Heritage Archive •But: Specification language complex, existing binaries hard to integrate •Prototype for portable binaries •Technologies: E.g., GNU Guix, Nix •Lightweight, isolated environments •Defined via command-line operations •Simpler specification, but less reproducible •Prototype for reproducible container builds •Technologies: E.g., Docker, Podman •Full system virtualizations •Created as snapshot of running system •No explicit specification: Defined via user actions •Requires virtualization platform •Technologies: E.g., VirtualBox, VMware •Detailed list of software dependencies •Not executable, but useful to compare environments •Critical for special hardware, e.g. on compute clusters •Technologies: E.g., EasyBuild CaseStudies WiKoDa App Isabelle Prover HPC-IO Traces REEs in the RDMC [1] Reusable Execution Environment Unpack content NFDI(xCS) Platform / Archives Download Download Reference References [1] Jan Bernoth et al. “Workflow for Creating and Sealing a Research Data Management Container(RDMC)”.In:SoftwareEngineering2025–CompanionProceedings.GI,2025. doi:10.18420/se2025-ws-26. [2] Sayash Kapoor and Arvind Narayanan. “Leakage and the reproducibility crisis in machine-learning-basedscience”.In:Patterns(2023).doi:10.1016/j.patter.2023. 100804. [3] Harald Semmelrock et al. “Reproducibility in Machine Learning-Driven Research”. In: CoRR (2023). doi:10.48550/arXiv.2307.10320. Funding The authors were funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under the National Research Data Infrastructure – NFDI 52/1 – 501930651.