オブジェクト指向並列化クラスライブラリの開発と性能評価

Translated title of the contribution: Development of Object-oriented Parallel Class Library and Performance Evaluation of Benchmark Code

小野 謙二, 玉木 剛, 野田 茂穂, 岩田 正子, 重谷 隆之

Research output: Contribution to journalArticle

Abstract

An Object-Oriented framework and class libraries with parallelism are designed and developed to enhance the construction of parallel physical simulators, which adopt domain decomposition method. Developed parallel data class is a class library that integrates a data structure and functions to manipulate the arrays for parallelisation. This data class can be used in both C/C++ and Fortran language. A benchmark code, which abstracts the part of a Jacobi iteration method for the elliptic partial differential equation, was ported using present framework and parallelized in C++/f90 mixed language. Measured timing results on Linux cluster showed us that the ported benchmark code using functions of the framework has almost the same performance and scalability of the original MPI benchmark code.
Original languageJapanese
Pages (from-to)44-53
Number of pages10
Journal情報処理学会論文誌コンピューティングシステム(ACS)
Volume48
Issue number8
Publication statusPublished - May 15 2007

Fingerprint

Domain decomposition methods
Partial differential equations
Data structures
Scalability
Simulators
Linux

Cite this

オブジェクト指向並列化クラスライブラリの開発と性能評価. / 小野謙二; 玉木剛; 野田茂穂; 岩田正子; 重谷隆之.

In: 情報処理学会論文誌コンピューティングシステム(ACS), Vol. 48, No. 8, 15.05.2007, p. 44-53.

Research output: Contribution to journalArticle

@article{b055e6456c024121aad93a6b5d5f0fd0,
title = "オブジェクト指向並列化クラスライブラリの開発と性能評価",
abstract = "An Object-Oriented framework and class libraries with parallelism are designed and developed to enhance the construction of parallel physical simulators, which adopt domain decomposition method. Developed parallel data class is a class library that integrates a data structure and functions to manipulate the arrays for parallelisation. This data class can be used in both C/C++ and Fortran language. A benchmark code, which abstracts the part of a Jacobi iteration method for the elliptic partial differential equation, was ported using present framework and parallelized in C++/f90 mixed language. Measured timing results on Linux cluster showed us that the ported benchmark code using functions of the framework has almost the same performance and scalability of the original MPI benchmark code.",
author = "謙二 小野 and 剛 玉木 and 茂穂 野田 and 正子 岩田 and 隆之 重谷",
year = "2007",
month = "5",
day = "15",
language = "Japanese",
volume = "48",
pages = "44--53",
journal = "情報処理学会論文誌コンピューティングシステム(ACS)",
issn = "1882-7829",
publisher = "一般社団法人情報処理学会",
number = "8",

}

TY - JOUR

T1 - オブジェクト指向並列化クラスライブラリの開発と性能評価

AU - 小野, 謙二

AU - 玉木, 剛

AU - 野田, 茂穂

AU - 岩田, 正子

AU - 重谷, 隆之

PY - 2007/5/15

Y1 - 2007/5/15

N2 - An Object-Oriented framework and class libraries with parallelism are designed and developed to enhance the construction of parallel physical simulators, which adopt domain decomposition method. Developed parallel data class is a class library that integrates a data structure and functions to manipulate the arrays for parallelisation. This data class can be used in both C/C++ and Fortran language. A benchmark code, which abstracts the part of a Jacobi iteration method for the elliptic partial differential equation, was ported using present framework and parallelized in C++/f90 mixed language. Measured timing results on Linux cluster showed us that the ported benchmark code using functions of the framework has almost the same performance and scalability of the original MPI benchmark code.

AB - An Object-Oriented framework and class libraries with parallelism are designed and developed to enhance the construction of parallel physical simulators, which adopt domain decomposition method. Developed parallel data class is a class library that integrates a data structure and functions to manipulate the arrays for parallelisation. This data class can be used in both C/C++ and Fortran language. A benchmark code, which abstracts the part of a Jacobi iteration method for the elliptic partial differential equation, was ported using present framework and parallelized in C++/f90 mixed language. Measured timing results on Linux cluster showed us that the ported benchmark code using functions of the framework has almost the same performance and scalability of the original MPI benchmark code.

M3 - 記事

VL - 48

SP - 44

EP - 53

JO - 情報処理学会論文誌コンピューティングシステム(ACS)

JF - 情報処理学会論文誌コンピューティングシステム(ACS)

SN - 1882-7829

IS - 8

ER -