Data Provenance and Reproducibility in Grid Based Scientific Workflows

One of the open issues in multi-physics simulation workflows is the collection and management of data provenance information, and the reproducibility of results. Simulation models interact by exchanging data, consequently data produced at each simulation step, depends on data produced in previous steps and exchanged data. In this paper we focus on collecting data provenance information based on this interaction of simulations. We introduce a non-intrusive way of collecting data provenance information, using WS-notifications produced by WSRF Services that expose data and simulation models. We generate the data dependency graph showing the "happened before" relationship between simulation models and data, in order to query data provenance information. We use semantics of simulation models and files stored in RDF format, provided by the data semantics system, to query data and interpret data provenance information. We demonstrate the feasibility of our data provenance system using the ServOSims Framework, a prototype platform for scientific workflows.