CAREER: Advanced Containers for Reproducibility in Computational and Data Science

可重复性对于科学进步和建立对科学成果的信任至关重要。已公布的计算结果越来越缺乏对配套信息的充分捕获和描述，从而使随后能够确认和扩展结果。该项目将设计和实施一种新的基于容器的方法来共享和复制科学成果。从该项目开发的可复制容器将跨不同的计算平台打包代码、数据、环境、出处和假设。与采用基于“DevOps”的方法不同，该方法使用户负担管理实验的再现性，而本项目使用科学实验的引用执行作为容器相关构件的一种虚拟化方法。虽然基于容器的方法可以帮助验证重复计算，但需要容器技术的进一步进步才能实现高级形式的重复性。该项目旨在实现可重复性，即使计算包括非确定性和竞争条件；代码、数据集和参数被更改；计算在分布式平台上执行；以及容器与敏感数据和未记录的内容共享。为此，该项目将开发一个开源容器运行时，该运行时将提供支持容器的可再运行性、可扩展性和发布能力的原语。这项工作利用了以前为计算科学开发的便携式容器。该奖项将为建立真实世界计算和数据科学用例的重现性奠定重要的基础。该项目将通过包括科学家、学生和教师在内的综合研究和教育计划，提高对计算可重复性工具的认识。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Reproducibility is essential for scientific progress and to establish trust in scientific results. Published computational results, increasingly, lack sufficient capture and description of companion information that enables subsequent confirmation and extension of results. This project will design and implement a novel container-based approach for sharing and reproducing scientific results. Reproducible containers developed from this project will package code, data, environment, provenance, and assumptions across heterogeneous computing platforms. In contrast to taking a "devops"-based approach, which burdens the user to manage reproducibility of experiments, this project uses reference executions of scientific experiments as a virtualization method for containerizing associated artifacts. While a container-based approach can help to verify repeat computations, further advancements in container technology are needed to enable advanced forms of reproducibility. This project aims to enable reproducibility even if computations include non-determinism and race conditions; code, data-sets, and parameters are changed; computations are performed on distributed platforms; and containers are shared with sensitive data and undocumented content. To that end, the project will develop an open-source container runtime that will offer primitives for enabling re-runnability, extensibility, and publish-ability of containers. The work leverages portable containers developed previously for computational sciences. This award will lay the foundation for an essential building block for establishing reproducibility of real-world computational and data science use cases. The project will increase awareness of the need for computational reproducibility tools through an integrated research and education plan involving scientists, students, and instructors.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.