Keywords
scientific workflow software, granularity, optimization
Start Date
1-7-2010 12:00 AM
Abstract
Scientific workflow management is an active area of research and developmentresponding to an increase in the complexity of computational models, data analysis anddata size. In the authors’ experience, the distinction between scientific workflow software(SWS) and modelling frameworks or toolsets is at best not clear in the minds of users ordevelopers, at least in the hydrology domain where the interest in concept of scientificworkflow appears rather recent. It is understandably tempting for some users to assumethat these new software tools aim to replace their existing modelling tools, with varyingexpectation depending on their prior satisfaction. While it is arguably clear that SWS canplay a role in improving practices for high-level orchestration and traceability of scientificworkflows, we explore in this paper the granularity at which activities can be usefullydefined. We describe a case study, the calibration of a model, wrapping the components ofa modelling framework (TIME) from the Trident and Kepler SWS. The problem isdecomposed in several workflows comprising activities of differing granularities. Weassess each approach against a set of criteria such as runtime performance and flexibility,discuss the feasibility and trade-off. The main findings are the design benefits stemmingfrom having to clearly identify separate activities in the process, and that the difficulty ofdecomposing the optimization problem into finer-grained activities increases mostmarkedly when needing iterative control flow capabilities.
On the appropriate granularity of activities in a scientific workflow applied to an optimization problem
Scientific workflow management is an active area of research and developmentresponding to an increase in the complexity of computational models, data analysis anddata size. In the authors’ experience, the distinction between scientific workflow software(SWS) and modelling frameworks or toolsets is at best not clear in the minds of users ordevelopers, at least in the hydrology domain where the interest in concept of scientificworkflow appears rather recent. It is understandably tempting for some users to assumethat these new software tools aim to replace their existing modelling tools, with varyingexpectation depending on their prior satisfaction. While it is arguably clear that SWS canplay a role in improving practices for high-level orchestration and traceability of scientificworkflows, we explore in this paper the granularity at which activities can be usefullydefined. We describe a case study, the calibration of a model, wrapping the components ofa modelling framework (TIME) from the Trident and Kepler SWS. The problem isdecomposed in several workflows comprising activities of differing granularities. Weassess each approach against a set of criteria such as runtime performance and flexibility,discuss the feasibility and trade-off. The main findings are the design benefits stemmingfrom having to clearly identify separate activities in the process, and that the difficulty ofdecomposing the optimization problem into finer-grained activities increases mostmarkedly when needing iterative control flow capabilities.
