79 巻 (2003) 8 号 p. 780-787
The experiences regarding the conctruction of PC clusters in three local laboratories are summarized. The number of CPU's varies from 2 to 32 (1 to 16 PC's). The performance of these clusters scales roughly with the number of CPU's. Using MPI, the merits of dual processing and/or hyper-threading technologies can be applied without any additional programming effort.