Virtual services in data grids

Data grids enable next generation scientific explorations that require intensive computation and analysis of petabyte-scale shared data collections. Apart from the challenge in creation and management of the data, another major challenge is the discovery of derived data products that have already been created. This work addresses the later challenge in minimizing response time and conserving the computing cycles in the grids by using "virtual services" to access derived data products. We analyze the role of virtual services for long running services in a service-oriented architecture and contribute a technique called "semantic mirror". A prototype was developed to discover derived data products of grid services using SRB-MCAT technology and GriPhyN Virtual Data Language (VDL).