@article{oai:ipsj.ixsq.nii.ac.jp:00018483,
 author = {片桐, 孝洋 and 吉瀬, 謙二 and 本多, 弘樹 and 弓場, 敏嗣 and Takahiro, Katagiri and Kenji, Kise and Hiroki, Honda and Toshitsugu, Yuba},
 issue = {SIG06(ACS6)},
 journal = {情報処理学会論文誌コンピューティングシステム（ACS）},
 month = {May},
 note = {本論文ではGram-Schmidt法（G-S法）を用いた再直交化処理の並列アルゴリズムにおいて，データ再分散を行うことで列方向分散（Column-Wise Distribution，CWD）における低い並列性を改善する再直交化方式を提案する．提案方式を用いた並列再直交化を，固有ベクトル計算のための逆反復法に実装した．PCクラスタおよび国産スーパコンピュータ3種（HITACHI SR8000/MPP，Fujitsu VPP800/63，およびNEC SX-5/128M8）を用いて提案方式が実装された並列逆反復法を評価したところ，CWDを用いた従来法に対し，修正G-S法を利用した再直交化で最大約4倍，古典G-S法を利用した再直交化で最大約3.5倍の速度向上が得られた．, In this paper, we improve a parallel re-orthogonalization with Gram-Schmidt (G-S) method. A data re-distribution approach is used to solve the low parallelism problem for column-wise distribution (CWD) in the method. The proposed method is implemented to the inverse iteration method for computing eigenvectors. The inverse iteration method using the proposed method is evaluated with a PC cluster and three kinds of super-computers in Japan, which are the HITACHI SR8000/MPP, Fujitsu VPP800/63, and NEC SX-5/128M8. The results of performance evaluation indicated that the maximum speedup factor of 4 in Modified G-S method, and of 3.5 in Classical G-S method with respect to conventional methods using CWD were obtained.},
 pages = {75--85},
 title = {データ再分散を行う並列Gram - Schmidt再直交化},
 volume = {45},
 year = {2004}
}