Auditing and Maintaining Provenance in Software Packages

Science projects are increasingly investing in computational reproducibility. Constructing software pipelines to demonstrate reproducibility is also becoming increasingly common. To aid the process of constructing pipelines, science project members often adopt reproducible methods and tools. CDE is a software packaging tool for reproducible research that encapsulates source code, datasets and environments. However, CDE is not amenable for creating software pipelines that demonstrate computational reproducibility. In this work, we propose software provenance to be included as part of CDE so that resulting provenanceincluded CDE packages can be easily used for creating software pipelines. We describe provenance attributes that must be included and how they can be efficiently stored in a light-weight CDE package. Further we show how a provenance in a package can be used for creating software pipelines and maintained as new packages are created. We experimentally evaluate the overhead of auditing and maintaining provenance and compare with heavy weight approaches for reproducibility such as virtualization. Our experiments indicate minimal overheads.