Storing and Discovering Critical Workflows from Log in Scientific Exploration

This paper aims at effectively reproducing the results of previous scientific workflow executions. We first identify the information that needs to be recorded in a workflow execution log, based on which we then determine the data flow among experiments. To effectively record a workflow execution log in a relational database management system (RDBMS) when the workflow design is absent, we propose a generic relational storage schema. Then techniques have been designed to automatically discover the minimal set of experiments that must be performed in order to reproduce a scientific result by posing appropriate SQL queries. Although such SQL queries can be evaluated using an off-the-shelf database system, we investigate the unique characteristics of the workflow log data and optimization techniques for evaluating such SQL queries efficiently.