A Dataflow-Oriented Atomicity and Provenance System for Pipelined Scientific Workflows

Scientific workflows have gained great momentum in recent years due to their critical roles in e-Science and cyberinfrastructure applications. However, some tasks of a scientific workflow might fail during execution. A domain scientist might require a region of a scientific workflow to be “atomic”. Data provenance, which determines the source data that are used to produce a data item, is also essential to scientific workflows. In this paper, we propose: (i) an architecture for scientific workflow management systems that supports both provenance and atomicity; (ii) a dataflow-oriented atomicity model that supports the notions of commit and abort; and (iii) a dataflow-oriented provenance model that, in addition to supporting existing provenance graphs and queries, also supports queries related to atomicity and failure.