Our proposed pilot project will explore the possibility of developing a more effective programming model for such distributed and data intensive environments, using an abstract mechanism for data management from the GridRPC community data handles. Analogous to the function handles that are central to the GridRPC paradigm, a data handle is an abstraction that can be dynamically bound to data and storage resources at run-time. A programmer can use a data handle as if it were variable, passing it between multiple GridSolve calls, while at the same time, the internal GridSolve run-time system can perform various operations (e.g. read, write) on them. We believe that if we can provide TGS with a data handle mechanism at the right level of abstraction, one that integrates different existing grid data management approaches while exposing the appropriate amount of structure to the scientific end user, then we can enable SCE-based TeraGrid applications that involve workflow patterns with a wide range complexity.