Flexible Scientific Workflows Using Frames and Dynamic Embedding
MetadataShow full metadata
Current approach to scientific workflow design in the popular open source Kepler system is based on the actor-oriented framework where concrete actors can be hierarchically composed and orchestrated by different directors (schedulers). A common assumption in this design framework is that workflow is static and must be completely specified before orchestration. Such a static and monolithic workflow cannot response to changing runtime conditions. We present flexible scientific workflow design that allows some tasks to be partially specified via abstract actors called Frame. The behavior of a frame is determined at runtime by the embedded concrete actor. We implemented the process of dynamic embedding that can tailor to different selection policies and enable automatic construction of subworkflow to execute the embedded component at runtime. Frames and dynamic embedding provide high level abstractions for specifying workflow that enables flexible execution nested to any level.