Optical Routing and Switching System

65076 High Performance Computing (HPC) clusters are constrained by the lack of adequate inter-processor communications bandwidth. Clusters built upon proprietary networks can be very expensive and difficult to support. This project will develop technology for constructing HPC clusters with very high bandwidth standards. The resulting clusters will be lower cost, higher performance, and more fault-tolerant. In Phase I, communications technology was surveyed, a Fibre Channel-based, switch-architecture, cluster platform was selected and designed, and a plan was developed for a low-cost, 200-CPU HPC cluster product. Phase II will design, develop, prototype, and test the processor nodes and the inter-processor communications structure. Existing cluster software tools will be re-hosted onto this cluster product. Commercial Applications and Other Benefits as described by the awardee: The proposed product targets the ¿sweet spot¿ range for HPC clusters, 100- to 200-CPU. It should be small enough to fit into someone¿s office, be low cost, and out-perform currently available high-end server clusters.