Effectiveness of Hybrid Workflow Systems for Computational Science

The workflow and its supporting systems are integral to computational science. Tailored to loosely coupled, and largely coarse-grained tasks, the workflow replaces the script as a way to automate the multiple steps of a large scale model. Workflow reuse has been at the subworkflow level but this restricts, over the long run, a workflow to running on the system on which it was developed. A scientist wanting to use two workflows developed by two different people and for different workflow systems will need to have access to both workflow systems. The contribution this paper makes is a qualitative and quantitative study of the tradeoffs of a hybrid workflow solution that utilizes multiple workflow systems and solutions to execute a single workflow. Our results indicate that the major tradeoffs are not in performance as much as they are in complexity.