Hypercube Project and a Simulator for a Hypercube of Computers

摘要

The Los Alamos National Laboratory and the University of New Mexico are working on HC, a hypercube project for large problems. A hypercube of 1024 nodes is being designed; implementation of a machine of at least 16 nodes is underway. The nodes of the machine are complete computer systems, including mass storage and a large main memory. The computational abilities of each node are enhanced by using a detached instruction fetch-execute-store organization. This allows the execution rate to achieve a consistently high level, approaching 90% of optimum. Trade-offs in the implementation methods indicate the speeds available with acceptable hardware investments. A simulator has been written which allows execution of high level programs on a hypercube organization. The simulator has shown that programs prepared for execution on a single stream machine can be easily modified to use the resources of a parallel machine, and that the results are identical to previous executions. 20 refs., 3 figs. (ERA citation 11:033471)