The Need for a Versioned Data Analysis Software Environment
This tackles the problem of costly and difficult reproducibility for researchers in high-energy physics and similar fields, though it is incremental as it builds on existing versioning and software distribution concepts.
The paper addresses the challenge of preserving complex data analysis runtime environments for reproducibility in scientific fields like high-energy physics, and presents a snapshotting file system optimized for software distribution, tested on environments such as those used in the Higgs boson discovery at the Large Hadron Collider.
Scientific results in high-energy physics and in many other fields often rely on complex software stacks. In order to support reproducibility and scrutiny of the results, it is good practice to use open source software and to cite software packages and versions. With ever-growing complexity of scientific software on one side and with IT life-cycles of only a few years on the other side, however, it turns out that despite source code availability the setup and the validation of a minimal usable analysis environment can easily become prohibitively expensive. We argue that there is a substantial gap between merely having access to versioned source code and the ability to create a data analysis runtime environment. In order to preserve all the different variants of the data analysis runtime environment, we developed a snapshotting file system optimized for software distribution. We report on our experience in preserving the analysis environment for high-energy physics such as the software landscape used to discover the Higgs boson at the Large Hadron Collider.