Parallelization of MCNP4 code by using simple FORTRAN algorithms

PI Yazid, M Takano, F Masukawa, Y Naito - 1993 - inis.iaea.org
PI Yazid, M Takano, F Masukawa, Y Naito
1993inis.iaea.org
[en] Simple FORTRAN algorithms, that rely only on open, close, read and write statements,
together with disk files and some UNIX commands have been applied to parallelization of
MCNP4. The code, named MCNPNFS, maintains almost all capabilities of MCNP4 in solving
shielding problems. It is able to perform parallel computing on a set of any UNIX
workstations connected by a network, regardless of the heterogeneity in hardware system,
provided that all processors produce a binary file in the same format. Further, it is confirmed …
[en] Simple FORTRAN algorithms, that rely only on open, close, read and write statements, together with disk files and some UNIX commands have been applied to parallelization of MCNP4. The code, named MCNPNFS, maintains almost all capabilities of MCNP4 in solving shielding problems. It is able to perform parallel computing on a set of any UNIX workstations connected by a network, regardless of the heterogeneity in hardware system, provided that all processors produce a binary file in the same format. Further, it is confirmed that MCNPNFS can be executed also on Monte-4 vector-parallel computer. MCNPNFS has been tested intensively by executing 5 photon-neutron benchmark problems, a spent fuel cask problem and 17 sample problems included in the original code package of MCNP4. Three different workstations, connected by a network, have been used to execute MCNPNFS in parallel. By measuring CPU time, the parallel efficiency is determined to be 58% to 99% and 86% in average. On Monte-4, MCNPNFS has been executed using 4 processors concurrently and has achieved the parallel efficiency of 79% in average.(author)
inis.iaea.org
以上显示的是最相近的搜索结果。 查看全部搜索结果