Evaluation of Deployment Options for a Graphical HPC Application

Author: Marvin Zein
Type: Bachelor's Thesis
Date: 2025-05-15
Reviewers: Prof. Dr. Michael Kuhn, Anna-Lena Roth
Supervisors: Anna-Lena Roth, Prof. Dr. Michael Kuhn
Download: PDF

Abstract

To make parallel programming education more accessible to students, the learning support tool EduMPI is being developed. EduMPI features a GUI that students can interact with to run parallel programs and analyze their performance on a HPC cluster. Currently, the GUI is available on Linux in the form of an AppImage. Even amongst professional developers, the most commonly used operating system is not Linux, but Windows. The GUI does not currently run on Windows and no working deployments exist. The goal of this thesis is to find a working deployment for Windows to make the GUI accessible for a wider range of students. The deployment solutions should be easy to deploy and maintain for developers and easy to use for students, all while keeping the functionality of the original AppImage deployment. Two different approaches for deploying the EduMPI GUI will be examined.
The first approach allows for changing the code base to create a native Windows deployment. Through resolving code and runtime related issues by making all parts of the code cross-compatible and setting up cross-compatible runtime dependencies, the GUI was made to work as expected on Windows. A manual deployment strategy and an automated deployment strategy for better developer experience using Github Actions were developed. The second approach makes use of virtualization and containerization by using the AppImage as it exists currently without making any changes to the GUI code base. Two deployment strategies using WSL2 and Docker were developed.
After comparing all deployment with each other, it was concluded that the total overall best deployment strategy in terms of functionality and user experience are the new native Windows and Linux deployments in combination with using Github Actions.