Sustainability and Reproducibility via Containerized Computing

Recent developments in the commercial open source community have catalysed the use of Linux containers for scalable deployment of web-based applications to the cloud. Scientific software can be containerized with dependencies, configuration files, post-processing tools and even simulation results, referred to as containerized computing. This new approach promises to significantly improve sustainability, productivity and reproducibility. We present our experiences, technology, and future plans for open source containerization of software used to model particle and radiation beams. Vagrant is central to our approach, using Docker for cloud deployment and VirtualBox virtual machines for deployment to Mac OS and Windows computers. Our technology enables seamless switching between the desktop and the cloud to simplify simulation development and execution.

💡 Research Summary

The paper presents a comprehensive case study on using container technologies to improve the sustainability, productivity, and reproducibility of scientific software that models particle and radiation beams. The authors begin by noting the rapid adoption of Linux containers in the commercial open‑source ecosystem, which has made it possible to deploy web‑based applications to the cloud with minimal friction. Building on this trend, they introduce the concept of “containerized computing,” wherein an entire scientific workflow—including the simulation engine, all required libraries, configuration files, post‑processing scripts, and even intermediate results—is packaged into a single container image.

Central to their approach is Vagrant, which acts as a meta‑orchestrator that can drive both Docker for cloud deployments and VirtualBox for desktop environments (macOS, Windows, and Linux). The Vagrantfile contains declarative specifications for Docker provisioning (using Dockerfile, docker‑compose, and related plugins) as well as VirtualBox VM configuration (memory, CPU, disk size, networking). By sharing a single Vagrantfile across environments, developers can write and debug code locally on a virtual machine, then seamlessly switch to a cloud instance where the same Docker image is pulled and executed at scale. This eliminates the classic “works on my machine” problem and provides a uniform execution environment regardless of the underlying host OS.

Technically, the authors construct a Dockerfile based on Ubuntu 20.04, layering scientific packages such as GEANT4, ROOT, Python, and CMake. They employ Vagrant plugins like vagrant-docker-compose and vagrant-disksize to define multi‑container topologies, persistent volumes, and network topology in a few lines of code. For cloud deployment, an AWS EC2 or Azure VM is provisioned with Docker Engine; Vagrant then automatically pulls the pre‑built image, creates the required containers, and starts the simulation workload. On a desktop, VirtualBox spins up a VM that mirrors the same environment, allowing developers to test changes without needing a cloud account.

Reproducibility is reinforced through immutable image digests and version‑tagged registries (Docker Hub or GitHub Packages). The paper demonstrates that any external researcher can clone the same Vagrantfile, pull the exact image by its digest, and obtain an identical computational environment. Moreover, the workflow is integrated with a CI/CD pipeline using GitHub Actions: on each push, the Docker image is rebuilt, unit‑ and integration‑tests are run (including a small benchmark simulation), and the image is pushed to the registry only if all tests pass. This automated gatekeeping ensures that published results can be regenerated at any time, satisfying the highest standards of scientific reproducibility.

From a sustainability perspective, the reliance on open‑source tools eliminates costly proprietary licenses, while containerization reduces hardware lock‑in. Containers are lightweight, enabling rapid snapshotting, version control, and rollback. Persistent data volumes are mounted to external storage (NFS, S3) so that large simulation outputs are not tied to a specific VM, facilitating data sharing and long‑term archiving. The authors also discuss future plans to migrate to Kubernetes for automated scaling of thousands of containers, and to adopt Helm charts for templated deployment of complex pipelines. They intend to embed FAIR‑compliant metadata (JSON‑LD) describing simulation parameters, software versions, and hardware configurations, thereby enhancing transparency and discoverability.

In conclusion, the combination of Vagrant, Docker, and VirtualBox provides a robust, declarative infrastructure that unifies development, testing, and production for beam‑physics simulations. It dramatically reduces the overhead of managing dependencies, improves developer productivity, guarantees that results can be reproduced by any collaborator, and offers a cost‑effective path toward long‑term software maintenance. The paper’s experience report and open‑source tooling serve as a valuable blueprint for other scientific domains seeking to modernize their computational workflows.