Model-Driven Approach to Workflow Specification in Scientific Computing David Woollard Ph.D. Candidate, Computer Science Department Advisor: Nenad Medvidovic University of Southern California, Los Angeles

Abstract

Scientists in a host of disciplines have begun utilizing computational technologies such as grid, cluster, and now cloud computing in order to conduct in silico experimentation. Scientific Workflows -- the composition of scientific tasks into stages of data processing, data dependencies between these stages, and deployment of stages onto computational resources -- has shown to be a very useful abstraction for the scientist conducting these experiments, though a number of significant engineering challenges remain before workflow specifications can become the schemata for conducting in silico science. Existing workflow models entwine scientific and engineering concerns, cannot be exchanged between workflow engine implementations, and only capture high-level data dependencies. Rather than directly interpreting workflow models, we advocate for a model-driven approach to workflow specification in which a high-level scientific workflow model be transformed into a domain-specific software architecture that can then be deployed onto multiple disparate computational platforms, including both Grid and Cloud solutions.

 

Bio

avid Woollard is a PhD Candidate at the University of Southern California. His interests are in the areas of software architecture and architectural recovery, specifically applied to the domain of scientific software. In addition to his work at USC, David is a full-time Software Engineer at NASA’s Jet Propulsion Laboratory, where he is a member of the Data Management Systems and Technologies Group.