Deciding How to Store Provenance

Provenance of a file is metadata pertaining to the history of the file. Provenance, unlike normal metadata stored in file systems, is retrieved primarily by running queries. This implies that provenance has to be indexed and should have a query interface. We believe that databases are the most appropriate place to store provenance as they provide both indexing and query capabilities. The goal of this paper is to explore the most appropriate schema and database technology for storing provenance. In the paper we discuss the different possible schemas for storing provenance and the tradeoffs in choosing each of the schemas. We then characterize the behavior of some of the popular database architectures under provenance recording/querying workloads. The database architectures that we considered are: RDBMS, Schemaless Embedded Databases (Berkeley DB), XML, and LDAP. Finally, we present preliminary performance results for the database architecture for provenance recording and some common provenance queries. Our results indicate that schemaless embedded databases have the best performance under most provenance workloads. The results also indicate that RDBMS has the best space utilization under most provenance workloads.