Experiences in using provenance to optimize the parallel execution of scientific workflows steered by users

The main advantages from using Scientific Workflow Management Systems to manage a large-scale scientific experiment are their automatic parallel execution and the improvement of result analysis through provenance data. Provenance data becomes especially useful for scientists when it is clearly associated to their domain data. In our experience, provenance data also reveals important optimizations opportunities in parallel execution and allows for user steering of workflows at run-time. The algebraic parallel execution engine is fine tuned by provenance statistics and users have explored provenance through steering support to visualize partial results from computational fluid dynamics simulations, to improve iterative uncertainty quantification applications in geophysics and to evaluate parameter setting and algorithms in several bioinformatics analyses. We discuss three of our real use cases of provenance data analysis with users from these different domains.