Distributed Numerical Weather Prediction via Satellite

This paper describes a recent undertaking in distributed numerical weather prediction via high data rate networks. The governing project involved the operation of a coupled mesoscale modeling system on widely separated supercomputers, and experiments and demonstrations were performed to explore both long-distance and local-area operation of the system. Connectivity was provided either by NASA's Advanced Communications Technology Satellite (ACTS) or by a terrestrial Asynchronous Transfer Mode (ATM) network. The modeling system was built around the Fifth-Generation Pennsylvania State University-National Center for Atmospheric Research Mesoscale Model (PSU-NCAR MM5) and consisted of the MM5, an ocean circulation model, and a surface wave model. The primary experiments and demonstrations utilized ACTS to link supercomputers at both NCAR in Boulder, Colorado, and at the Ohio Supercomputer Center in Columbus, Ohio. Other experiments and demonstrations utilized an ATM network to link supercomputers running the system at sites in Boulder. This paper focuses on the design and performance of the distributed numerical weather prediction system and its supporting networks. The project has demonstrated the feasibility of remote, distributed supercomputing, the potential for application of distributed computing principles and message-passing software to numerical modeling and coupled model construction, and means for creatively applying state-of-the-art networking technology to support meteorological modeling.