Seamless Coarse Grained Parallelism Integration in Intensive Bioinformatics Workflows

To be easily constructed, shared and maintained, complex in silico bioinformatics analysis are structured as workflows. Furthermore, the growth of computational power and storage demand from this domain, requires workflows to be efficiently executed. However, workflow performances usually rely on the ability of the designer to extract potential parallelism. But atomic bioinformatics tasks do not often exhibit direct parallelism which may appears later in the workflow design process. In this paper, we propose a Model-Driven Architecture approach for capturing the complete design process of bioinformatics workflows. More precisely, two workflow models are specified: the first one, called design model, graphically captures a low throughput prototype. The second one, called execution model, specifies multiple levels of coarse grained parallelism. The execution model is automatically generated from the design model using annotation derived from the EDAM ontology. These annotations describe the data types connecting differents elementary tasks. The execution model can then be interpreted by a workflow engine and executed on hardware having intensive computation facility