Network-Enabled Solvers and the NetSolve Project

The beginning of the 21st century will present new challenges for largescale applications involving communication with, and coordination of, large numbers of geographically dispersed information sources, supplying information to a large number of geographically dispersed information consumers. Applications of this class will require an environment which supports longterm or continuous, reliable and fault-tolerant, highly distributed, heterogeneous, and scalable computing capability. The characteristics of such applications include: distributed data collection, distributed computation (often in signi cant amounts), distributed control, and distributed output. Many of these applications will require high reliability and continuous operation, even though individual nodes or links may fail or otherwise be unavailable. Such applications will be constructed out of a wide variety of computational components (including smart sensors, personal digital assistants, workstations, and supercomputers), and a wide variety of communications media (wire, optical ber, terrestrial radio, satellite) with varying degrees of link reliability, bandwidth, and message loss. The reliability requirement means that such applications must degrade gracefully rather than fail in the presence of node or link failures, or with insu cient communications bandwidth and high message loss rates. Since some computational resources may not be available on a continuous basis, applications may have to adapt to varying computational power. The potential for hostile attack to such systems requires that they have a high degree of security, both for authentication of data and privacy of sensitive information.