Prov-Dominoes: an exploratory analysis approach for provenance data

Abstract

AbstractProvenance, the record of the history of a piece of information, has become increasinglyrelevant to understanding, auditing, and reproducing computational tasks. The prove-nance analysis processes can often be overwhelming to the user due to the large volume ofdata, the multiple relationships among data, and the implicit information buried into thedata. Some existing tools provide provenance analysis support based on node-link dia-grams, relying on visualization features over edges and vertices. Others are workflow-basedtools, such as VisTrails and Taverna. However, none of them support the exploration ofimplicit provenance data, such as the inferences of the PROV Data Model Constraints.In this work, we introduce Prov-Dominoes, a tool designed to explore provenance datainteractively. Prov-Dominoes promotes the provenance relationships among entities, ac-tivities, and agents into first-class elements, represented by domino tiles. Moreover, itallows users to combine and compose such domino tiles visually and interactively, us-ing GPU. We evaluated Prov-Dominoes over distinct case studies, helping us to observeProv-Dominoes in action. We were able to uncover implicit relationships in a dataset ofanimal characteristics, identify the parameters that influenced workflow execution results,and highlight essential activities in a smart home. We also evaluated the performance ofsequential combinations executed in Prov-Dominoes when dealing with provenance datawith thousands of relations, contrasting their executions in GPU and CPU. The resultsshowed that, for a large data set, GPU outperformed CPU by two orders of magnitude.