DataArchive

How to set up an ATS model-data archive

This best-practices document is based on discussion of a team of ATS users and developers, and builds on concepts of replicable and reproducible computation (e.g. Heroux 2019, SIAM News). DOE and most other funding sources are strengthening their requirements on model-data archiving, as are journals and the scientific modeling community of practice as a whole. This aims to set up a directory structure and conceptual model for how to run hypothesis-driven numerical experiments using ATS.

Our goals in this are to:

1. Ensure all needed scripts and input datasets are provided either in this location or in well-documented, archival quality data repositories.
2. Ensure sufficient documentation is provided so that a different user (or the same user at a later time) could execute the same experiment for some significant period of time in the future (of the order 5-10 years) and reasonably expect to get the same or very similar results (limited by availability of specific libraries or comparable hardware, but not by source code availability or dataset availability).
3. Ensure sufficient traceability so that a different user could execute a similar experiment on their own data, enabling the community to build on previous efforts.
4. Do so in a sufficiently standardized way across the ATS user community to limit the difficulty of doing #2 or #3.

NOTE: this is a Work in Progress!