Grid Computing Research Laboratory State University of New York (SUNY) Binghamton Department of Computer Science

Wei Lu, Kenneth Chiu, Satoshi Shirasuna, and Dennis Gannon
"A Hybrid Decomposition Scheme for Building Scientific Workflows",
to appere in Proceedings of High Performance Computing Symposium (HPC 2007),
Norfolk, Virginia, March 25-29, 2007
[PDF] [bibtex]

Abstract
Two problem decomposition schemes, component assembly and workflow orchestration, have been widely adopted to architect large scale scientific applications. These two methodologies, however, approach problem decomposition from different distinctly perspectives with the result that most problem solving environments provide only one approach to the exclusion of the other. For example the Ccaffeine, a parallel CCA framework, provides the component assembly environment only; while Kepler, a widely-used scientific workflow toolkit, is designed mainly as the work- flow orchestration environment. Each methodology has situations, within the same problem domain, where it may be more appropriate than the other, however. Thus, to bring benefits from both methodologies, in this paper we present a hybrid problem decomposition scheme. By augmenting Ccaffeine with the web services interface, we enable Ccaffeine as a special workflow actor in the Kepler environment, so that a Kepler user can gain the benefits of both approaches by applying the two methodologies for the subproblems depending on the various performance and resource-sharing requirements. The hybrid scheme will first use the workflow scheme to decompose the problem based on the distribution of the resource; thereafter adopts the component assembly scheme to further decompose those computationally intensive cores for the high performance solutions.

Key Words