SGProv: Summarization Mechanism for Multiple Provenance Graphs

Abstract

Scientific workflow management systems (SWfMS) are powerful tools in the automation of scientific experiments. Several workflow executions are necessary to accomplish one scientific experiment. Data provenance, typically collected by SWfMS during workflow execution, is important to understand, reproduce and analyze scientific experiments. Provenance is about data derivation, thus it is typically represented in the form of a directed acyclic graph. For each workflow execution, a provenance graph is generated. Numerous graphs are generated after several workflow runs, exploring different parameters. The resulting provenance database requires considerable storage space and querying it involves handling a large volume of graphs. Typical provenance queries process many graphs to get data derivation paths (lineage). This article proposes SGProv, a summarization mechanism for provenance graphs, using a graph database to store and query them. The goal is to generate a single small summary graph that represents all provenance graphs generated during an experiment, eliminating redundant data. This summarization approach aims to reduce the processing time of provenance queries by using only the summary graph to answer them without the need for rebuilding the original graphs. Results of provenance queries on the summary graph, from typical workflow executions, show performance improvements without data loss on query results.