Coping with Cycles in Provenance

This paper considers the issue of cycles in provenance metadata. Cyclic provenance poses a problem as they hide the distinction between ancestors and decendants. Yet this distinction is at the heart of provenance. Our contributions are: development of terminology, defining the parameters of our ideal solution and present and evaluate four approaches to cyclic provenance. Before diving into a set of goals and possible solutions, we define terms for discussing cyclic provenance. This terminology is then used to state our objective. Informally, our ideal approach would avoid on disk cycles, retain maximal information using minimal data. We then present and compare five approaches to coping with cyclic provenance: retaining cycles, versioning, node merging, provenance copying and implicit versioning. We describe the concept behind each approach, summarize its implementation and assess its strengths and weaknesses.