"Computational Grids" are comprised of distributed computational sites, remote instruments, data archives, networks, and other resources. Grids are becoming an increasingly prevalent and important platform for high-performance computing. However application performance on the Grid is often difficult to achieve due to the dynamic and heterogeneous performance characteristics of the underlying Grid resources.
A powerful paradigm for achieving application performance on the Grid is adaptivity. The goal of the AppleS (Application-Level Scheduler) Project is to develop a methodology and software for applying adaptivity to the scheduling and execution of Grid applications. AppLeS applications demonstrate that adaptivity is an effective and powerful paradigm for achieving application performance in the most difficult of environments - where multiple users share resources, resources exhibit distinct and dynamic performance characteristics, and computation, communication and data must be coordinated over local and wide-area networks. In this talk, we describe the methodology, prototype software, and new projects developed as part of the AppLeS project.