초록 ▼

A parallel computing system and method having improved performance where a program is concurrently run on a plurality of nodes for reducing total processing time, each node having a processor, a memory, and a predetermined number of communication channels connected to the node and independently conn

A parallel computing system and method having improved performance where a program is concurrently run on a plurality of nodes for reducing total processing time, each node having a processor, a memory, and a predetermined number of communication channels connected to the node and independently connected directly to other nodes. The present invention improves performance of performance of the parallel computing system by providing a system which can provide efficient communication between the processors and between the system and input and output devices. A method is also disclosed which can locate defective nodes with the computing system.

대표청구항 ▼

[ We claim:] [1.] A method for parallel processing, comprising:(a) inputting data and a problem-solving algorithm for a problem to be solved by a N-dimension parallel computer using logarithmic fanout;(b) configuring a sufficient subspace of dimension m required to solve the problem, where m is less

[ We claim:] [1.] A method for parallel processing, comprising:(a) inputting data and a problem-solving algorithm for a problem to be solved by a N-dimension parallel computer using logarithmic fanout;(b) configuring a sufficient subspace of dimension m required to solve the problem, where m is less than or equal to N;(c) mapping a n-dimension spatial grid, where n is less than or equal to m, by allocating work to specific processors of the parallel computer;(d) opening the subspace and inputting the data, execution parameters, and the algorithm into the subspace using logarithmic fanout;(e) processing data according to algorithm within each processor;(f) writing messages to each of the selected neighboring processors paired in directions in a specified order of writing first to the neighboring processor in a fore orientation and then writing to the neighboring processor in an aft orientation such that the startup, copy, and transfer time for a subsequent message overlaps the transfer time for a previous message;(g) repeating step (f) until all selected neighboring processors of the n-dimensional spatial grid have been written to;(h) and, when at a boundary, and in the direction of a boundary where there is only one orientation, writing to only one processor associated with that orientation;(i) subsequently reading messages from each of the selected neighboring processors in paired directions in the same order of directions, but reading first from the neighboring processor in the aft orientation and then reading from the neighboring processor in the fore orientation such that the startup and transfer time for a subsequent message overlaps the transfer and copy time for a previous message;(j) repeating step (i) until all selected neighboring processors of the n-dimensional spatial grid have been read; and(k) and, when at a boundary, and in the direction of a boundary where there is only one orientation, reading from only one processor associated with that orientation;(l) outputting the solution to the problem;(m) closing the subspace; and(n) displaying the solution to the problem.