This project will investigate a community service system architecture and run-time infrastructure needed to support such high-end services. We will focus on two key run-time issues for performance: dynamic resource management and scaling. Dynamic resource management allows: (i) resources to be {de-} allocated to service requests on-the-fly, (ii) resources to be acquired via negotiation, and (iii) resources to be shared across inter-service domains. Resource management will be based on a performance estimation technique that learns service execution time as function of dependent inputs and resources allocated. We will produce a set of next generation tools and libraries that can be leveraged by community services running in single-site clusters and in multi-site Grids. The toolset will be evaluated by applying it to several high-end network services in the area of visualization, astronomy, and bioinformatics.