Issues in reconfigurability and adaptability in heterogeneous distributed systems for high-performance computing are the focus of the work funded by this grant. Our efforts are part of an ongoing research project in metacomputing and are a follow on to the DOE funded PVM system that has witnessed over a decade of use at numerous institutions worldwide. The current project, termed Harness, investigates novel methodologies and tools for distributed metacomputing, focusing on dynamically reconfigurable software frameworks. During the first phase, we defined the metacomputing architecture embodied in Harness and developed prototype subsystems as proof of concept exercises. Subsequently, we designed and developed a complete software framework manifesting the Harness architecture, and also developed several tools and subsystems that demonstrated the viability and effectiveness of our proposed model for next generation metacomputing. We then used this substrate to emulate multiple programming environments on Harness, and conducted performance evaluation and tuning exercises. The main research results from these efforts include the establishment of software metacomputing systems as viable and cost-effective alternatives to MPPs; the demonstration of dynamic and reconfigurable platforms as effective methods of tailoring parallel computing environments; the development of methodologies to construct plugin modules for component-based distributed systems; contributions to performance modeling and optimization in emulated software environments; and software architectures for multi- and mixed-paradigm parallel distributed computing.
PDF
download
878987797
028-85220240
