| 摘要 |
A multicomputer system and method for automatic sequential-to-parallel program partition, scheduling and execution. The plurality of computers (100, 101, 102, 103) are connected via a uni-directional slotted ring (120) (backbone). The ring (120) supports, with deterministic delays, point-to-point, broadcast and EXCLUSIVE-READ operations over labeled tuples. The execution of parallel programs uses a connectionless computing method that forms customized coarse grain SIMD, MIMD and pipelined processors automatically using the connected multicomputers. The disclosed multicomputer architecture is also resilient to processor and backbone failures. |
