Matchmaking support for dynamic workflow composition

Service description and discovery offer complementary challenges, but in both cases, the problem is finding the right trade-off between accuracy and generality that will result in a positive service identification. Discovery systems have historically tended to focus on domain-specific techniques using single sources of knowledge to help classify queries against services, making both maintenance and extension difficult. The primary contribution of this paper is the presentation of a generic brokerage framework based on the use of plug-in components, that are themselves web services. The framework has been developed in the context of the KNOOGLE project, where the focus has been on demonstrating support for (i) the discovery of Grid services for the GridSAM job submission system and (ii) integration with the Taverna workflow enactment system. However the broker itself is domain independent and it is the multiple user-specified matchmaker plug-ins that act as sources of domain-specific knowledge. The broker collects the results of the matchmakers' comparison of the query and service and then applies a user-specified selection policy to determine the final choice of service. Thus a range of comprehensive packaging of brokerage functionality becomes possible through the use of supplied and user-defined matchers and supplied or user-defined selection policies.