Abstract:

After a dcsor;p;.:on of the parallel computation model for the class of large-scale commuting problems in the field of fluid dynamics?a specialized supercomputer architecture "which realizes massive parallelism is proposed in 'his paper. The architecture proposed is a personal supercomputer consisting oi a host system and an array oi computing nodes.The architecture achieves reduced complexity, increased parallelism by taking advantage of the special purpose feature of the system. The interconnection topology reflects the locality property of data exchanges (luring computation, Knch computer node itself has a pipelined vector processor containing two independent pipelines. Thus the system combines the parallelism between computing nodes at a higher level with the vector processing capability at a lower level to increase computing power significantly. The special purpose feature and the existence of the host reduce the software complexity. The issues in hardware, software and programming are discussed and the communication performance is studied and evaluated in the paper.

Key words: supercomputer, fluid dynamics, parallel processing, hyperring, pipelined architecture