[Paper Review] DUNE as an Example of Sustainable Open Source Scientific Software Development

Markus Blatt|arXiv (Cornell University)|Sep 6, 2013

Distributed and Parallel Computing Systems3 references3 citations

TL;DR

This paper presents DUNE as a sustainable open-source scientific software framework for solving partial differential equations, emphasizing modular design, community-driven development, and collaboration between academia and industry. By employing a semi-open development model, rigorous code reviews, and a flexible licensing scheme, DUNE achieves long-term maintainability and broad adoption across diverse scientific domains.

ABSTRACT

In this paper we describe how DUNE, an open source scientific software framework, is developed. Having a sustainable software framework for the solution of partial differential equations is the main driver of DUNE's development. We take a look how DUNE strives to stay sustainable software.

Motivation & Objective

• To address the unsustainable development practices in scientific software, where legacy code is extended without maintainability or documentation.
• To create a unified, modular, and extensible interface for solving partial differential equations across diverse numerical methods and applications.
• To establish a sustainable software ecosystem through open development, community involvement, and industrial collaboration.
• To ensure long-term maintainability and usability by enforcing code reviews, stable release cycles, and backward compatibility.
• To enable researchers to avoid reimplementing core functionality by providing a well-documented, reusable, and extensible software framework.

Proposed method

• Adopting a modular architecture with seven core modules and support for user-contributed modules, each with its own build system.
• Implementing a semi-open development model with public mailing lists, regular developer and user meetings, and voting rights for core developers.
• Using version control (Subversion until 2013, then Git) and public patch review via mailing lists to ensure code quality and interface consistency.
• Introducing a runtime exception to the GPL-2 license to allow integration into proprietary software, increasing adoption by industry.
• Establishing a formal proposal and review process for interface changes, requiring consensus among core developers.
• Encouraging industrial participation through contract development and shared infrastructure contributions, reducing burden on academic developers.

Experimental results

Research questions

• RQ1 How can open-source scientific software achieve long-term sustainability in the face of evolving research needs and institutional constraints?
• RQ2 What development model best supports modularity, reusability, and extensibility in scientific software for partial differential equations?
• RQ3 How can collaboration between academic researchers and industry partners be effectively structured to enhance software quality and maintainability?
• RQ4 What role do public code reviews and community involvement play in improving software stability and reducing technical debt?
• RQ5 How can licensing models support both open collaboration and industrial adoption in scientific software frameworks?

Key findings

• DUNE’s modular architecture and use of modern C++ templates allow high performance while enabling fine-grained access to data structures without sacrificing efficiency.
• The semi-open development model—with public discussions, regular meetings, and voting rights—ensures that changes are well-justified and widely accepted by the community.
• The inclusion of a runtime exception in the GPL-2 license significantly increased industrial adoption, with major companies using and even open-sourcing their DUNE-based tools.
• Regular, predictable releases (approximately twice a year) help users focus on stable codebases while still benefiting from recent features.
• Experienced users actively contribute bug fixes and testing, effectively serving as human testers and improving software stability.
• Industrial contributions have reduced the maintenance burden on academic developers by funding infrastructure work, testing, and usability improvements.