●分子科学計算推進室
設置経緯と役割について
昨年4月に、大学共同利用機関法人 自然科学研究機構 分子科学研究所から情報基盤センターに赴任いたしました南部です。どうぞよろしくお願いいたします。分子研時代は、もっぱら分子科学計算のためのスーパーコンピュータ環境の運用および関連した科学の研究を行って参りました。そこで、今まで培った経験を生かし活用する場として、様々な方の助言を頂き、情報基盤センター内に分子科学計算推進室の設置を行うことが出来ました。この場をお借りして、お礼を申し上げます。
さて、この分子科学計算推進室とは「何ぞや？」となるのですが、近年、利用者の多様化とともに計算環境も多様化しさらには、一人の研究者では把握しきれない分野と分野の境目のような境界領域に新たな科学が出現しています。生物と情報、物理と化学、生物と化学、生物と物理という境界領域です。あるいは、一対一ではなく、生物と情報と化学と物理という輪が、分子というキーワードで結びついているように感じられます。まるで、DNAのように、分子がらせん構造を持ちながら繋がっているように、複雑です。そのため、ある現象のメカニズムを把握するために突如、分子科学計算が必要になってしまっているような利用者が見受けられます。そこで、そのような研究をある意味、共同研究も含めて、専門分野の視点からサポートしようという試みを始めました。つまり、「分子科学」分野に関連するソフトウェアの開発及びアプリケーションプログラムの個別サポートを積極的に行うというのが主な趣旨です。
ロドプシン
昨年4月に、大学共同利用機関法人 自然科学研究機構 分子科学研究所から情報基盤センターに赴任いたしました南部です。どうぞよろしくお願いいたします。分子研時代は、もっぱら分子科学計算のためのスーパーコンピュータ環境の運用および関連した科学の研究を行って参りました。そこで、今まで培った経験を生かし活用する場として、様々な方の助言を頂き、情報基盤センター内に分子科学計算推進室の設置を行うことが出来ました。この場をお借りして、お礼を申し上げます。
さて、この分子科学計算推進室とは「何ぞや？」となるのですが、近年、利用者の多様化とともに計算環境も多様化しさらには、一人の研究者では把握しきれない分野と分野の境目のような境界領域に新たな科学が出現しています。生物と情報、物理と化学、生物と化学、生物と物理という境界領域です。あるいは、一対一ではなく、生物と情報と化学と物理という輪が、分子というキーワードで結びついているように感じられます。まるで、DNAのように、分子がらせん構造を持ちながら繋がっているように、複雑です。そのため、ある現象のメカニズムを把握するために突如、分子科学計算が必要になってしまっているような利用者が見受けられます。そこで、そのような研究をある意味、共同研究も含めて、専門分野の視点からサポートしようという試みを始めました。つまり、「分子科学」分野に関連するソフトウェアの開発及びアプリケーションプログラムの個別サポートを積極的に行うというのが主な趣旨です。
ロドプシン関係各位へ
いつもお世話になっております。
情報基盤センター主催の講演会についてご案内させていただきます。
ご参加下さいますようお願い申し上げます。（特に，学生さんの参加を歓迎いたします。）
九州大学
情報基盤センター
南部伸孝
記
講師，Michael W. Schmidt教授
Iowa State University
演題：”Parallelization of the GAMESS electronic structure program”
日時：6月22日（木）午後3時より
場所：情報基盤センター多目的講習室（SCS室）（３階）
福岡市東区箱崎６－１０－１（箱崎キャンパス）
要旨：
The General Atomic and Molecular Electronic Structure System (GAMESS) program is a large FORTRAN program for quantum chemical and related computations. Quantum chemistry is characterized as seeking high accuracy solutions but at the expense of considerable machine resources: disk, memory, and processor time. Algorithms scale in their CPU requirements from O(N**4) to O(N**7), with the ubiquitous repulsion integrals and density matrices requiring O(N**4) storage. This requires storage of data outside main memory, traditionally on disk, but more recently in the memory of the other nodes in a parallel computer. The main focus of the talk will be about the effort of more than ten years to enable efficient parallel execution of GAMESS, rather than on quantum chemistry itself.
The Distributed Data Interface (DDI) is a library distributed with GAMESS, for the purpose of sharing memory between nodes. Today's ideal hardware consists of 64 bit processors, probably in a 4-way SMP node, with a large local memory, and therefore an enormous remote memory, accessed by a high bandwidth low latency network. DDI relies on SystemV memory calls within a SMP node, and TCP/IP, MPI, or vendor libraries (such as Cray's SHMEM) to carry all underlying messages between nodes. Conversion of a purely serial GAMESS code in 1992 to a largely parallel GAMESS+DDI code at the present time, has involved parallelization of existing quantum chemistry codes, development of parallel codes from scratch, and teaching students how to do parallel coding. Recently two new sophisticated pieces of GAMESS were developed as distributed memory, parallel codes only - a sure sign of evolving away from the serial CPU. Some performance examples will be given for existing methods. All of these technologies are accessible to ordinary chemistry departments on modest budgets, e.g. dozens of processor in clusters priced near 1,000,000 yen, not just high-end supercomputers.
Concluding remarks will outline future challenges at the high end, in the anticipation these too will be relevant five years into the future: multicore or multithreaded processors, system board architectures, the role of disks, or how to use multi-thousand CPU systems.
以上
Dear Colleagues,
We will have Open Lecture on Thursday, June 22. Everyone is welcome.
Date & Time: June 22 (Thursday), 15:00
Place: SCS Room (3rd floor) Computing and Communications Center Build. Kyushu Univ.
Lecturer: Professor Michael W. Schmidt
Affiliation: Iowa State University
Title:
Parallelization of the GAMESS electronic structure program
Abstract:
The General Atomic and Molecular Electronic Structure System (GAMESS) program is a large FORTRAN program for quantum chemical and related computations. Quantum chemistry is characterized as seeking high accuracy solutions but at the expense of considerable machine resources: disk, memory, and processor time. Algorithms scale in their CPU requirements from O(N**4) to O(N**7), with the ubiquitous repulsion integrals and density matrices requiring O(N**4) storage. This requires storage of data outside main memory, traditionally on disk, but more recently in the memory of the other nodes in a parallel computer. The main focus of the talk will be about the effort of more than ten years to enable efficient parallel execution of GAMESS, rather than on quantum chemistry itself.
The Distributed Data Interface (DDI) is a library distributed with GAMESS, for the purpose of sharing memory between nodes. Today's ideal hardware consists of 64 bit processors, probably in a 4-way SMP node, with a large local memory, and therefore an enormous remote memory, accessed by a high bandwidth low latency network. DDI relies on SystemV memory calls within a SMP node, and TCP/IP, MPI, or vendor libraries (such as Cray's SHMEM) to carry all underlying messages between nodes. Conversion of a purely serial GAMESS code in 1992 to a largely parallel GAMESS+DDI code at the present time, has involved parallelization of existing quantum chemistry codes, development of parallel codes from scratch, and teaching students how to do parallel coding. Recently two new sophisticated pieces of GAMESS were developed as distributed memory, parallel codes only - a sure sign of evolving away from the serial CPU. Some performance examples will be given for existing methods. All of these technologies are accessible to ordinary chemistry departments on modest budgets, e.g. dozens of processor in clusters priced near 1,000,000 yen, not just high-end supercomputers.
Concluding remarks will outline future challenges at the high end, in the anticipation these too will be relevant five years into the future: multicore or multithreaded processors, system board architectures, the role of disks, or how to use multi-thousand CPU systems.
Best wishes, Shinkoh NANBU.
無事，終了いたしました。参加された方へこころから感謝をいたします。
次回は，7月11，12日の二日間にわたって，分子動力学プログラムAmber入力作成及び実行と量子化学計算プログラムGaussianにおけるPCM計算のコツについて，国立オーストラリア大学から2名の講師をお呼びして行う予定です。ご参加ください。
いつもお世話になっております。
情報基盤センター主催の講演会についてご案内させていただきます。
ご参加下さいますようお願い申し上げます。（特に，学生さんの参加を歓迎いたします。）
九州大学
情報基盤センター
南部伸孝
記
講師，Michael W. Schmidt教授
Iowa State University
演題：”Parallelization of the GAMESS electronic structure program”
日時：6月22日（木）午後3時より
場所：情報基盤センター多目的講習室（SCS室）（３階）
福岡市東区箱崎６－１０－１（箱崎キャンパス）
要旨：
The General Atomic and Molecular Electronic Structure System (GAMESS) program is a large FORTRAN program for quantum chemical and related computations. Quantum chemistry is characterized as seeking high accuracy solutions but at the expense of considerable machine resources: disk, memory, and processor time. Algorithms scale in their CPU requirements from O(N**4) to O(N**7), with the ubiquitous repulsion integrals and density matrices requiring O(N**4) storage. This requires storage of data outside main memory, traditionally on disk, but more recently in the memory of the other nodes in a parallel computer. The main focus of the talk will be about the effort of more than ten years to enable efficient parallel execution of GAMESS, rather than on quantum chemistry itself.
The Distributed Data Interface (DDI) is a library distributed with GAMESS, for the purpose of sharing memory between nodes. Today's ideal hardware consists of 64 bit processors, probably in a 4-way SMP node, with a large local memory, and therefore an enormous remote memory, accessed by a high bandwidth low latency network. DDI relies on SystemV memory calls within a SMP node, and TCP/IP, MPI, or vendor libraries (such as Cray's SHMEM) to carry all underlying messages between nodes. Conversion of a purely serial GAMESS code in 1992 to a largely parallel GAMESS+DDI code at the present time, has involved parallelization of existing quantum chemistry codes, development of parallel codes from scratch, and teaching students how to do parallel coding. Recently two new sophisticated pieces of GAMESS were developed as distributed memory, parallel codes only - a sure sign of evolving away from the serial CPU. Some performance examples will be given for existing methods. All of these technologies are accessible to ordinary chemistry departments on modest budgets, e.g. dozens of processor in clusters priced near 1,000,000 yen, not just high-end supercomputers.
Concluding remarks will outline future challenges at the high end, in the anticipation these too will be relevant five years into the future: multicore or multithreaded processors, system board architectures, the role of disks, or how to use multi-thousand CPU systems.
以上
Dear Colleagues,
We will have Open Lecture on Thursday, June 22. Everyone is welcome.
Date & Time: June 22 (Thursday), 15:00
Place: SCS Room (3rd floor) Computing and Communications Center Build. Kyushu Univ.
Lecturer: Professor Michael W. Schmidt
Affiliation: Iowa State University
Title:
Parallelization of the GAMESS electronic structure program
Abstract:
The General Atomic and Molecular Electronic Structure System (GAMESS) program is a large FORTRAN program for quantum chemical and related computations. Quantum chemistry is characterized as seeking high accuracy solutions but at the expense of considerable machine resources: disk, memory, and processor time. Algorithms scale in their CPU requirements from O(N**4) to O(N**7), with the ubiquitous repulsion integrals and density matrices requiring O(N**4) storage. This requires storage of data outside main memory, traditionally on disk, but more recently in the memory of the other nodes in a parallel computer. The main focus of the talk will be about the effort of more than ten years to enable efficient parallel execution of GAMESS, rather than on quantum chemistry itself.
The Distributed Data Interface (DDI) is a library distributed with GAMESS, for the purpose of sharing memory between nodes. Today's ideal hardware consists of 64 bit processors, probably in a 4-way SMP node, with a large local memory, and therefore an enormous remote memory, accessed by a high bandwidth low latency network. DDI relies on SystemV memory calls within a SMP node, and TCP/IP, MPI, or vendor libraries (such as Cray's SHMEM) to carry all underlying messages between nodes. Conversion of a purely serial GAMESS code in 1992 to a largely parallel GAMESS+DDI code at the present time, has involved parallelization of existing quantum chemistry codes, development of parallel codes from scratch, and teaching students how to do parallel coding. Recently two new sophisticated pieces of GAMESS were developed as distributed memory, parallel codes only - a sure sign of evolving away from the serial CPU. Some performance examples will be given for existing methods. All of these technologies are accessible to ordinary chemistry departments on modest budgets, e.g. dozens of processor in clusters priced near 1,000,000 yen, not just high-end supercomputers.
Concluding remarks will outline future challenges at the high end, in the anticipation these too will be relevant five years into the future: multicore or multithreaded processors, system board architectures, the role of disks, or how to use multi-thousand CPU systems.
Best wishes, Shinkoh NANBU.
無事，終了いたしました。参加された方へこころから感謝をいたします。
次回は，7月11，12日の二日間にわたって，分子動力学プログラムAmber入力作成及び実行と量子化学計算プログラムGaussianにおけるPCM計算のコツについて，国立オーストラリア大学から2名の講師をお呼びして行う予定です。ご参加ください。
