Decoupling Provenance Capture and Analysis from Execution

Capturing provenance usually involves the direct observation and instrumentation of the execution of a program or workflow. However, this approach restricts provenance analysis to pre-determined programs and methods. This may not pose a problem when one is interested in the provenance of a well-defined workflow, but may limit the analysis of unstructured processes such as interactive desktop computing. In this paper, we present a new approach to capturing provenance based on full execution record and replay. Our approach leverages full-system execution trace logging and replay, which allows the complete decoupling of analysis from the original execution. This enables the selective analysis of the execution using progressively heavier instrumentation.