Nonintrusive collection and management of data provenance in scientific workflows
University:Univ. of Athens
Time:4:00pm (coffee: 3:30)
We introduce an efficient mechanism to collect, store, and retrieve data provenance information in workflows of multiphysics simulations. Using notifications, we enable the non-intrusive collection of information about workflow events during workflow execution. Combining these events with workflow structure information, constant for every execution of a workflow, we obtain the data provenance information for the specific run of the workflow. Data provenance information is structured into a graph that represents workflow events on the basis of their causal dependency. We use a graph database to store this graph and utilize the traversal framework provided, to efficiently retrieve data provenance information from the graph by traversing backwards from a data object to every workflow event that is part of its provenance. Finally, we integrate data provenance information with semantics of workflow services to provide complete and meaningful data provenance information.