ホーム >第4回JCAHPCセミナー


最先端共同HPC基盤施設(JCAHPC: Joint Center for Advanced High Performance Computing, http://www.jcahpc.jp/)は,筑波大学計算科学研究センターと東京大学情報基盤センターとが共同で設立した組織です。国内最高性能を有するOakforest-PACSシステムはJCAHPCによって設計,導入,運用されています(HPL性能13.55 PFLOPS,2017年6月発表のTOP 500リスト(http://www.top500.org)では世界第7位,国内第1位)。


第4回目となる今回は,数値ライブラリ分野の研究者として世界的に著名なMichael Heroux博士(Senior Scientist, Sandia National Laboratories, USA)をお招きして,Heroux博士がリーダーをつとめるTrillinosプロジェクト(https://trilinos.org/)の現状と動向のほか,Heroux博士が中心となって整備を進め,新しいスパコンベンチマークとして注目されているHPCG(http://hpcg-benchmark.org/)についてもご講演いただきます。そのほか,筑波大・東大両センターの教員から関連した講演があります。講演は英語で実施されます。




会場:東京大学情報基盤センター(浅野) 4階 413遠隔会議室

 遠隔配信(柏会場):東京大学柏キャンパス第2総合研究棟 3階 315会議室2

 遠隔配信(つくば会場):筑波大学計算科学研究センター ワークショップ室



参加費:無料 事前登録等は不要です


4th JCAHPC Seminar
13:30-17:30, August 29 (Tuesday), 2017
Remote Conference Room (4F 413), Information Technology Center (ITC), The University of Tokyo

13:30 - 13:35  Taisuke Boku (JCAHPC/University of Tsukuba)
Welcome & Opening
13:35 - 14:35 Mike Heroux (Sandia National Laboratories, USA)
The Trilinos Project Exascale Roadmap [Abstruct]
14:35 - 15:05 Akira Imakura (JCAHPC/University of Tsukuba)
A complex moment-based parallel eigensolver using the block communication-avoiding Arnoldi procedure [Abstruct]
15:05 - 15:35 Tetsuya Sakurai (JCAHPC/University of Tsukuba)
Nonlinear Sakurai-Sugiura method for complex band structure calculation on Oakforest-PACS [Abstruct]
15:55 - 16:25 Mike Heroux (Sandia National Laboratories, USA)
The High Performance Conjugate Gradients (HPCG) Benchmark [Abstruct]
16:25 - 16:55 Kengo Nakajima (JCAHPC/University of Tokyo)
ppOpen-HPC and Sparse Linear Solvers [Abstruct]
16:55 - 17:25 Akihiro Ida (JCAHPC/University of Tokyo)
Efficient Structures of H-matrices on Distributed Memory Computer Systems [Abstruct]
17:25 - 17:30 Kengo Nakajima (JCAHPC/University of Tokyo)


〒113-8658 東京都文京区弥生2−11−16
東京大学 情報基盤センター



Title: The Trilinos Project Exascale Roadmap

Abstract: Preparing for Exascale computer systems requires fundamental changes in algorithms, software design and implementation. While effective use of on-node concurrency has been essential for some years, new algorithm classes are still emerging, and the design and implementation of reusable software libraries for these new platforms is still unsettled. In this presentation, we describe the Exascale roadmap for the Trilinos project, a collection of reusable scientific software libraries for constructing, managing and solving large scientific computing problems on parallel systems. We discuss our efforts to expose and utilize new concurrency resources through changes in algorithms, design choices and implementation strategies for reusable libraries, as application codes move to expose greater concurrency and asynchronous execution.

Title: A complex moment-based parallel eigensolver using the block communication-avoiding Arnoldi procedure

Abstract: For solving interior eigenvalue problems, complex moment-based eigensolvers have been actively studied because of their high parallel efficiency. In this talk, we introduce some typical complex moment-based eigensolvers and analyzed their relationship. We also propose a novel method using a communication-avoiding Arnoldi procedure (s-step Arnoldi procedure). We evaluate the performance of the proposed method and compare with other complex moment-based eigensolvers.

Title: Nonlinear Sakurai-Sugiura method for complex band structure calculation on Oakforest-PACS

Abstract: Electronic transport calculations require the computation of eigenvalues located in a ring region and their corresponding eigenvectors of a quadratic eigenvalue problem with T-palindromic structure. In this talk, we propose a technique for solving such quadratic problems by using the nonlinear Sakurai-Sugiura method. The proposed method reduces computational costs using a T-palindromic structure and achieves high parallel efficiency by using a contour integral. The performance of the proposed method is evaluated on Oakforest-PACS.

Title: The High Performance Conjugate Gradients (HPCG) Benchmark

Abstract: The High Performance Conjugate Gradients (HPCG) Benchmark is a new community metric for ranking high performance computing systems. The first list of results was released at ISC'14, including optimized results for systems built upon Fujitsu, Intel, Nvidia technologies. Presently the list contains more than 100 entries, including the majority of the top 50 platforms, and HPCG has been accepted into the TOP500 ranking system. In this presentation we discuss the design of HPCG, opportunities for optimizing its performance, show the most recent results, and discuss possible future efforts.

Title: ppOpen-HPC and Sparse Linear Solvers

Abstract: "ppOpen-HPC" is an open source infrastructure for development and execution of optimized and reliable simulation code on post-peta-scale (pp) parallel computers based on many-core architectures, such as the Oakforest-PACS system of Joint Center for Advanced High Performance Computing (JCAHPC). "ppOpen-HPC" is part of a five-year project (FY.2011-2015) spawned by the "Development of System Software Technologies for Post-Peta Scale High Performance Computing" funded by JST-CREST. The framework covers various types of procedures for scientific computations in various types of computational models, such as FEM, FDM, FVM, BEM and DEM. Automatic tuning (AT) technology enables automatic generation of optimized libraries and applications under various types of environments. We release the most updated version of ppOpen-HPC as open source software every year in November (2012-2015), which is available at http://ppopenhpc.cc.u-tokyo.ac.jp/ppopenhpc/ . In 2016, the team of ppOpen-HPC joined ESSEX-II (Equipping Sparse Solvers for Exascale) project (Leading P.I. Professor Gerhard Wellein (University of Erlangen-Nuremberg)), which is funded by JST-CREST and the German DFG priority programme 1648 "Software for Exascale Computing" (SPPEXA) under Japan (JST)-Germany (DFG) collaboration until FY.2018. In ESSEX-II, we develop pK-Open-HPC (extended version of ppOpen-HPC, framework for exa-feasible applications), preconditioned iterative solvers for quantum sciences, and a framework for automatic tuning (AT) with performance model. In the presentation, various types of achievements of ppOpen-HPC, ESSEX-II, and pK-OpenHPC project, focusing on parallel preconditioned iterative solvers will be shown with future perspective.

Title: Efficient Structures of H-matrices on Distributed Memory Computer Systems

Abstract: Thehierarchicalmatrices (H-matrices) are one of approximation techniques for dense matrices appearing in scientific applications. Any H-matrix can be transformed to an H-matrix with block low-rank representation (BLR). Although the BLR is easy to perform matrix arithmeticcompared with the normal H-matrix, it ends up increasing memory usage from O() of the normal H-matrix. For this reason, the BLR has been utilized for complex arithmetic of relatively small matrices on a CPU node. In this presentation, we discuss the applicability of the H-matrices withBLR for simple arithmetic, such as the H-matrix-vector multiplication, in large scale problems on distributed memory computer systems.


2013年3月,筑波大学と東京大学で締結された「計算科学・工学及びその推進のための計算機科学・工学の発展に資するための連携・協力推進に関する協定」に基づき,筑波大学計算科学研究センターと東京大学情報基盤センターは,「最先端共同HPC基盤施設(JCAHPC: Joint Center for Advanced High Performance Computing)」を設置しました。JCAHPCは東大情報基盤センタースーパーコンピュータシステム柏拠点(東大柏キャンパス)に両機関の教職員が中心となり設計するスーパーコンピュータシステムを設置し,最先端の大規模高性能計算基盤を構築・運営するための組織です。既にアナウンスされているように,JCAHPCでは本年12月1日よりOakforest-PACSシステム(OFP)の運用を開始します。OFPはIntel Xeon Phi 7250(Knights Landing, KNL)を8,208ノードを搭載し,ピーク性能は25 PFLOPS,運用開始時点で国内最大・最速のシステムとなるものと考えられます。

両センターは本施設を連携・協力して運営することにより,最先端の計算科学を推進し,我が国の学術及び科学技術の振興に寄与していく所存です。 その一環として,国内外の研究者による「JCAHPCセミナー」を開催していく予定です。