Ninf-1/Ninf-G を用いたNMR蛋白質立体構造決定のための遺伝アルゴリズムのグリッド化  [in Japanese] Gridifying a Genetic Algorithm for NMR Three-dimensional Protein Structure Determination by Using Ninf-1 and Ninf-G  [in Japanese]

Access this Article

Search this Article

Author(s)

Abstract

本論文では,Ono らが提案したNMR 蛋白質立体構造決定のための遺伝アルゴリズム(GA)を,複数のWAN 上のPC クラスタ群から構成されるグリッド上で並列化したシステムを提案し,提案システムの性能評価を行った結果を報告する.提案システムは,マスタ,サブマスタ,ワーカから構成される階層的なマスタ・ワーカ方式を用いて並列化されている.マスタと各PC クラスタ間の通信はセキュアなGridRPC ミドルウェアNinf-G を用いて,また,PC クラスタ内の通信は高速なGridRPCミドルウェアNinf-1 を用いて実現されている.さらに,提案システムでは,Ninf-G によるインターネット上の通信遅延を隠蔽するため,スライド転送手法を導入している.5 サイト/1 196CPU から構成されるグリッドテストベッドで,78 残基からなる蛋白質の立体構造決定問題を用いて,提案システムの性能評価を行った結果,高い並列化効率を示すことが確認された.In this paper, we parallelize the genetic algorithm (GA) for NMR protein three-dimensional structure determination, which has been proposed by Ono et al., on a grid that consists of multiple PC clusters on the WAN and report some results on the performance evaluation of the proposed system. The proposed system is parallelized with the hierarchical master-worker paradigm and consists of a master, submasters and workers. The communication between the master and each PC cluster is realized with Ninf-G, which is a secure GridRPC middleware, and that in each PC cluster is implemented by using Ninf-1, which is a fast GridRPC middleware. In the proposed system, we employ the slide transfer technique in order to hide the latency of communication on the Internet by using Ninf-G. The experimental results on the grid testbed consisting of 5 sites/1,196 CPUs showed that the proposed system effectively utilized computing resources on the grid testbed when it was applied to a problem of determining the three-dimensional structure of a 78-residue protein.

In this paper, we parallelize the genetic algorithm (GA) for NMR protein three-dimensional structure determination, which has been proposed by Ono et al., on a grid that consists of multiple PC clusters on the WAN and report some results on the performance evaluation of the proposed system. The proposed system is parallelized with the hierarchical master-worker paradigm and consists of a master, submasters and workers. The communication between the master and each PC cluster is realized with Ninf-G, which is a secure GridRPC middleware, and that in each PC cluster is implemented by using Ninf-1, which is a fast GridRPC middleware. In the proposed system, we employ the slide transfer technique in order to hide the latency of communication on the Internet by using Ninf-G. The experimental results on the grid testbed consisting of 5 sites/1,196 CPUs showed that the proposed system effectively utilized computing resources on the grid testbed when it was applied to a problem of determining the three-dimensional structure of a 78-residue protein.

Journal

  • 情報処理学会論文誌コンピューティングシステム(ACS)

    情報処理学会論文誌コンピューティングシステム(ACS) 46(SIG12(ACS11)), 396-406, 2005-08-15

    Information Processing Society of Japan (IPSJ)

Cited by:  3

  • Grid-Oriented Genetic Algorithms for Large-Scale Optimization  [in Japanese]

    ONO Isao

    計測と制御 = Journal of the Society of Instrument and Control Engineers 47(6), 473-479, 2008-06-10

    References (35) Cited by (1)

  • Construction and Operation of the Grid Challenge Testbed  [in Japanese]

    AIDA KENTO , OSAWA KIYOSHI , OSUMI TOMOTAKA , KASAI TAKEFUMI , ONO ISAO , JITSUMOTO HIDEYUKI , MATSUOKA SATOSHI , SAITO HIDEO , ENDO TOSHIO , YOKOYAMA DAISAKU , TAURA KENJIRO , CHIKAYAMA TAKASHI , TANAKA YOSHIO , SHIMOSAKA HISASHI , KAJIWARA HIROKI , HIROYASU TOMOYUKI , FUJISAWA KATSUKI

    IPSJ SIG Notes 107, 49-54, 2006-08-02

    References (19) Cited by (3)

  • メインストリームを目指すHPC  [in Japanese]

    廣安 知之

    [O]perations research as a management science [r]esearch 52(10), 645-650, 2007-10-01

    References (12)

Codes

  • NII Article ID (NAID)
    110002769855
  • NII NACSIS-CAT ID (NCID)
    AA11833852
  • Text Lang
    JPN
  • Article Type
    Journal Article
  • ISSN
    1882-7829
  • NDL Article ID
    7463874
  • NDL Call No.
    Z74-C192
  • Data Source
    CJPref  NDL  NII-ELS  IPSJ 
Page Top