Abstract
A parallel mapping algorithm for plasma particle codes on vector-parallel computers is shown with the use of an extended Abe-Nishihara vectorization algorithm. It is scalable not only for performance but also for size of problem. A plasma particle code employing this algorithm achieves scalable parallel performance on the distributed-memory-type vector-parallel computer VPP300 and the SX-4, the shared- and distributed-memory-type vector-parallel architecture. On such a code, parallel performance is determined by four ratios. First is a ratio between the number of arithmetic operations and the number of elements which are to be transferred to other processors. Second is a peak ratio between the arithmetic flop/s and communication MB/s performance. Third is a ratio between efficiency of the arithmetic and communication performance. Fourth is a ratio of the number of meshes to particles. Among those four ratios, the three ratios except for the third one are determined uniquely by the property of the code or computer. Therefore, the third ratio is an important issue to design and select parallel computers for the plasma particle codes.
