Looking at Software Sustainability and Productivity Challenges from NSF

This paper is a contribution to the Computational Science & Engineering Software Sustainability and Productivity Challenges (CSESSP Challenges) Workshop (https://www.nitrd.gov/csessp/), sponsored by the Networking and Information Technology Research and Development (NITRD) Software Design and Productivity (SDP) Coordinating Group, held October 15th-16th 2015 in Washington DC, USA. It introduces the role of software at the National Science Foundation (NSF) and the NSF Software Infrastructure for Sustained Innovation (SI2) program, then describes challenges that the SI2 program has identified, including funding models, career paths, incentives, training, interdisciplinary work, portability, and dissemination, as well as lesson that have been learned.

💡 Research Summary

The paper presents a concise yet comprehensive overview of the challenges facing software sustainability and productivity as identified by the National Science Foundation’s Software Infrastructure for Sustained Innovation (SI2) program. It begins by contextualizing the pivotal role that software now plays in modern scientific and engineering research, positioning the SI2 initiative as a strategic effort to fund and nurture long‑lived, reusable, and community‑driven software assets. Despite early successes, the authors argue that the program has surfaced a set of systemic obstacles that hinder the creation of a robust, self‑reinforcing software ecosystem.

Seven interrelated challenge categories are enumerated. First, the funding model remains project‑centric and short‑term, providing little assurance for the ongoing maintenance, bug fixing, and incremental improvement that mature software requires. Second, career pathways for software‑focused researchers are ill‑defined; traditional academic promotion criteria prioritize publications over code contributions, leading to talent attrition. Third, the incentive structure fails to recognize software development as a scholarly output, discouraging researchers from investing significant effort in engineering best practices. Fourth, there is a training gap: curricula rarely cover software engineering principles, reproducibility, or sustainability, leaving domain scientists and software engineers with mismatched skill sets. Fifth, interdisciplinary collaboration is hampered by cultural and organizational silos, making it difficult to align domain expertise with engineering rigor. Sixth, portability across heterogeneous computing environments—cloud, high‑performance clusters, edge devices—remains a technical hurdle, often resulting in code that is locked to a specific platform. Finally, dissemination and adoption suffer from inconsistent licensing, lack of standardized metadata, and fragmented distribution channels, limiting reuse and reproducibility.

For each challenge, the authors distill lessons learned from SI2’s experience. They advocate for dedicated “maintenance” grant streams that can support long‑term stewardship of critical software, and for the creation of permanent research software engineer positions within universities and national labs. They propose revising tenure and promotion guidelines to formally credit software artifacts, and suggest that funding agencies adopt metrics that capture code quality, community uptake, and impact. In education, they call for the development of modular, cross‑disciplinary curricula that blend domain science with software engineering, supplemented by workshops, bootcamps, and mentorship programs. To foster interdisciplinary work, they recommend clear role definitions in collaborative agreements and the establishment of joint appointment mechanisms. On the technical front, the paper urges the adoption of containerization, modular architectures, and community‑driven standards to improve portability and ease of integration. Regarding dissemination, the authors stress the importance of open‑source licensing, the use of persistent repositories (e.g., GitHub, Zenodo), and the implementation of rich, machine‑readable metadata to enhance discoverability.

The conclusion emphasizes that addressing these five pillars—funding, career structures, incentives, training, and technology—requires coordinated policy reforms at the NSF and broader research community level. Without such systemic changes, the United States risks losing the long‑term benefits of its substantial investment in scientific software, potentially eroding its competitive edge in computational science. The paper thus serves both as a diagnostic report and a roadmap for building a sustainable software ecosystem that can keep pace with the accelerating demands of modern research.