PARALLEL PROCESSING METHOD FOR LU DECOMPOSITION FOR MEMORY DISTRIBUTED PARALLEL COMPUTER SYSTEM COMPRISING SMP NODE

摘要

In a memory distributed parallel computer in which a plurality of SMP nodes are interconnected by a network, each of the SMP nodes is composed of a plurality of processors, a memory module, and a cache. Each of the processors is composed of a plurality of arithmetic units called cores. If one node is composed of 64 cores, one core is assigned for communication use and the other 63 cores are assigned for computation use. While the cores for computation use perform LU decomposition, the core for communication use transfers data the computation of which is completed and which is necessary for the data update of the other nodes.