• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to project page

2015 Fiscal Year Annual Research Report

10億並列・エクサスケールスーパーコンピュータの耐故障性基盤

Research Project

Project/Area Number 23220003
Research InstitutionTokyo Institute of Technology

Principal Investigator

松岡 聡  東京工業大学, 学術国際情報センター, 教授 (20221583)

Co-Investigator(Kenkyū-buntansha) 實本 英之  東京工業大学, 学術国際情報センター, 助教 (00545311)
Project Period (FY) 2011-04-01 – 2016-03-31
Keywordsハイパフォーマンスコンピューティング / エクサスケールコンピューティング / 耐故障性技術 / データ圧縮 / チェックポイント・リスタート / バーストバッファ
Outline of Annual Research Achievements

本年度は、1)これまで開発してきた手法の統合に向けた技術開発、2)更なる障害復旧アルゴリズム・システムの改良を行った。

1) システム統合の一環として、これまでに提案してきた統合環境FMIを含むチェックポイント手法、耐故障に関する数理モデリングの検証を可能とする、故障発生器の構築を行った。この故障発生器は、仮想化技術を用いており、従来のアプリケーションプロセスを対象としたものに加え、ハードウェアやOSに対して故障を挿入することが可能となり、計算環境全体を対象とした統合的なモデル検証を実現する。現時点では、主な故障種としてメモリ故障を対象とした実装を行っており、これまで対象としていたfail-stop な故障モデルに加え、メモリデータ化けによる誤った計算結果の出力(SDC: Silent Data Corruption)の再現も可能となっている。

また、2) 耐故障機能の改良については、非可逆圧縮によるチェックポイント手法について、科学技術計算に特化した、ウェーブレット変換を用いた非可逆圧縮手法を提案した。これは、可逆圧縮の効率が落ちる浮動小数に対応する為の手法であり、チェックポイントに誤差を含んでしまう一方、高圧縮を実現できる。このため、元データの精度や計算結果の用途によって、許容可能な誤差閾値を定めることによって、チェックポイントを効率よく圧縮し、チェックポイント保存時間の削減が可能となる。実気象アプリケーション(NICAM) を利用した評価では、誤差1.2%の状態で、圧縮サイズは一般のgzipに対し80%程度となった。

Research Progress Status

27年度が最終年度であるため、記入しない。

Strategy for Future Research Activity

27年度が最終年度であるため、記入しない。

  • Research Products

    (55 results)

All 2016 2015 Other

All Int'l Joint Research (3 results) Journal Article (29 results) (of which Int'l Joint Research: 18 results,  Peer Reviewed: 23 results,  Open Access: 29 results) Presentation (22 results) (of which Int'l Joint Research: 20 results,  Invited: 22 results) Book (1 results)

  • [Int'l Joint Research] Lawrence Livermore National Laboratory/Oak Ridge National Laboratory(米国)

    • Country Name
      U.S.A.
    • Counterpart Institution
      Lawrence Livermore National Laboratory/Oak Ridge National Laboratory
  • [Int'l Joint Research] ZIH, TU Dresden(ドイツ)

    • Country Name
      GERMANY
    • Counterpart Institution
      ZIH, TU Dresden
  • [Int'l Joint Research] Shanghai Supercomputer Center, SJTU(中国)

    • Country Name
      CHINA
    • Counterpart Institution
      Shanghai Supercomputer Center, SJTU
  • [Journal Article] Towards Understanding the Performance of FPGAs using OpenCL Benchmarks2016

    • Author(s)
      Hamid Reza Zohouri, Naoya Maruyama, Aaron Smith, Motohiko Matsuda, Satoshi Matsuoka
    • Journal Title

      HiPEAC 2016

      Volume: - Pages: -

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Design of a NVRAM Specialized Degree Aware Dynamic Graph Data Structure2016

    • Author(s)
      Keita Iwabuchi, Roger Pearce, Brian Van Essen, Maya Gokhale, Satoshi Matsuoka
    • Journal Title

      7th Annual Non-Volatile Memories Workshop 2016

      Volume: - Pages: -

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Towards a Distributed Large-Scale Dynamic Graph Data Store2016

    • Author(s)
      Keita Iwabuchi, Roger Pearce, Brian Van Essen, Maya Gokhale, Satoshi Matsuoka
    • Journal Title

      Graph Algorithms Building Blocks (GABB’2016)

      Volume: - Pages: -

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Routing on the Dependency Graph: A New Approach to Deadlock-Free High-Performance Routing2016

    • Author(s)
      Jens Domke, Torsten Hoefler, Satoshi Matsuoka
    • Journal Title

      HPDC'16

      Volume: - Pages: -

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Adaptive Multi-level Blocking Optimization for Sparse Matrix Vector Multiplication on GPU2016

    • Author(s)
      Yusuke Nagasaka, Akira Nukada, Satoshi Matsuoka
    • Journal Title

      International Conference on Computational Science (ICCS 2016)

      Volume: - Pages: -

    • Peer Reviewed / Open Access
  • [Journal Article] Understanding Performance Portability of OpenACC for Supercomputers2015

    • Author(s)
      Suttinee Sawadsitang, James Lin, Simon See, Francois Bodin, Satoshi Matsuoka
    • Journal Title

      Programming Models, Languages and Compilers Workshop for Manycore and Heterogeneous Architectures (PLC2015) in conjunction with IPDPS2015

      Volume: - Pages: 699 - 707

    • DOI

      10.1109/IPDPSW.2015.60

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Modeling Gather and Scatter with Hardware Performance Counters for Xeon Phi2015

    • Author(s)
      James Lin, Akira Nukada, Satoshi Matsuoka
    • Journal Title

      The 15th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID 2015) Doctoral Symposium

      Volume: - Pages: 713 - 716

    • DOI

      10.1109/CCGrid.2015.59

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Characterizing MPI and Hybrid MPI+Threads Applications at Scale: Case Study with BFS2015

    • Author(s)
      Abdelhalim Amer, Huiwei Lu, Pavan Balaji, Satoshi Matsuoka
    • Journal Title

      The 2nd Workshop on Parallel Programming Model for the Masses (PPMM 2015) in conjunction with the 15th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID 2015)

      Volume: - Pages: 1075 - 1083

    • DOI

      10.1109/CCGrid.2015.93

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] GPUクラスタにおける大規模都市気流シミュレーションの最適化と性能モデル2015

    • Author(s)
      高嵜 祐樹,遠藤 敏夫,松岡 聡
    • Journal Title

      ハイパフォーマンスコンピューティングと計算科学シンポジウム論文集

      Volume: 2015 Pages: 129 - 137

    • Peer Reviewed / Open Access
  • [Journal Article] Performance Analysis of MapReduce Implementations for High Performance Homology Search2015

    • Author(s)
      Chaojie Zhang, Koichi Shirahata, Shuji Suzuki, Yutaka Akiyama, Satoshi Matsuoka
    • Journal Title

      ハイパフォーマンスコンピューティングと計算科学シンポジウム論文集

      Volume: 2015 Pages: 73 - 80

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Efficient Execution of Multiple CUDA Applications using Transparent Suspend, Resume and Migration2015

    • Author(s)
      Taichiro Suzuki, Akira Nukada, Satoshi Matsuoka
    • Journal Title

      Euro-Par 2015 Parallel Processing

      Volume: 9233 Pages: 687-699

    • DOI

      10.1007/978-3-662-48096-0_53

    • Peer Reviewed / Open Access
  • [Journal Article] Signal-Driven Swarming: A Parallel Implementation of Evolved Autonomous Agents to Perform A Foraging Task2015

    • Author(s)
      Aleksandr Drozd, Olaf Witkowski, Satoshi Matsuoka, Takashi Ikegami
    • Journal Title

      The First International Symposium on Swarm Behavior and Bio-Inspired Robotics

      Volume: - Pages: -

    • DOI

      10.13140/RG.2.1.1022.1920

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Discovering Aspectual Classes of Russian Verbs in Untagged Large Corpora2015

    • Author(s)
      Aleksandr Drozd, Anna Gladkova, Satoshi Matsuoka
    • Journal Title

      IEEE DSDIS 2015

      Volume: - Pages: 61 - 68

    • DOI

      10.1109/DSDIS.2015.30

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Python, Performance, and Natural Language Processing2015

    • Author(s)
      Aleksandr Drozd, Anna Gladkova, Satoshi Matsuoka
    • Journal Title

      the 5th Workshop on Python for High-Performance and Scientific Computing @ SC15

      Volume: - Pages: -

    • DOI

      10.1145/2835857.2835858

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Evaluating AVX2 Vgather Instruction with Stencils2015

    • Author(s)
      James Lin, Qiang Qin, Shuo Li, Minhua Wen and Satoshi Matsuoka
    • Journal Title

      HPC China 2015

      Volume: - Pages: -

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Hardware-Centric Analysis of Network Performance for MPI Applications2015

    • Author(s)
      Kevin Brown, Satoshi Matsuoka and Jens Domke
    • Journal Title

      21st IEEE International Conference on Parallel and Distributed Systems (ICPADS 2015)

      Volume: - Pages: 692 - 699

    • DOI

      10.1109/ICPADS.2015.92

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] mrCUDA: A middleware for migrating rCUDA virtual GPUs to native GPUs2015

    • Author(s)
      Pak Markthub, Akihiro Nomura, Satoshi Matsuoka
    • Journal Title

      情報処理学会 研究報告ハイパフォーマンスコンピューティング(HPC)

      Volume: 2015-HPC-150 (6) Pages: 1-9

    • Open Access
  • [Journal Article] ノード内同時実行ジョブにおけるパフォーマンスカウンタによるプロセス毎消費電力のモデル化2015

    • Author(s)
      寺西 賢人,野村 哲弘,遠藤 敏夫,松岡 聡
    • Journal Title

      情報処理学会 研究報告ハイパフォーマンスコンピューティング(HPC)

      Volume: 2015-HPC-150 (28) Pages: 1-6

    • Open Access
  • [Journal Article] Cloud-based Burst Buffers for I/O Acceleration2015

    • Author(s)
      Tianqi Xu, Kento Sato, Satoshi Matsuoka
    • Journal Title

      情報処理学会 研究報告ハイパフォーマンスコンピューティング(HPC)

      Volume: 2015-HPC-150 (20) Pages: 1-11

    • Open Access / Int'l Joint Research
  • [Journal Article] OpenCLによるFPGAの予備評価2015

    • Author(s)
      丸山直也, Hamid Reza Zohouri, 松田元彦, 松岡聡
    • Journal Title

      情報処理学会 研究報告ハイパフォーマンスコンピューティング(HPC)

      Volume: 2015-HPC-150 (44) Pages: 1-5

    • Open Access
  • [Journal Article] 疎行列ベクトル積計算を対象としたGPU向けメモリアクセス削減手法2015

    • Author(s)
      長坂侑亮, 額田彰, 松岡聡
    • Journal Title

      情報処理学会 研究報告ハイパフォーマンスコンピューティング(HPC)

      Volume: 2015-HPC-151 (8) Pages: 1-7

    • Open Access
  • [Journal Article] Optimizing the Rodinia Benchmark for FPGAs2015

    • Author(s)
      Hamid Reza Zohouri, Naoya Maruyama, Aaron Smith, Motohiko Matsuda, Satoshi Matsuoka
    • Journal Title

      情報処理学会 研究報告ハイパフォーマンスコンピューティング(HPC)

      Volume: 2015-HPC-152 (16) Pages: 1-7

    • Open Access / Int'l Joint Research
  • [Journal Article] Towards Cloud-based Burst Buffers for I/O Intensive Computing in Cloud (poster)2015

    • Author(s)
      Tianqi Xu, Kento Sato, Satoshi Matsuoka
    • Journal Title

      HPC in Asia Workshop in conjunction with the International Supercomputing Conference (ISC’15)

      Volume: - Pages: -

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] mrCUDA: Low-Overhead Middleware for Transparently Migrating CUDA Execution from Remote to Local GPUs (poster)2015

    • Author(s)
      Pak Markthub, Akihiro Nomura, Satoshi Matsuoka
    • Journal Title

      Supercomputing2015 (SC15)

      Volume: - Pages: -

    • Peer Reviewed / Open Access
  • [Journal Article] Design of a NVRAM Specialized Dynamic Graph Data Structure (poster)2015

    • Author(s)
      Keita Iwabuchi, Roger Pearce, Brian Van Essen, Maya Gokhale, Satoshi Matsuoka
    • Journal Title

      Supercomputing2015 (SC15)

      Volume: - Pages: -

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] Multi-level Blocking Optimization for Fast Sparse Matrix Vector Multiplication on GPUs (poster)2015

    • Author(s)
      Yusuke Nagasaka, Akira Nukada, Satoshi Matsuoka
    • Journal Title

      Supercomputing2015 (SC15)

      Volume: - Pages: -

    • Peer Reviewed / Open Access
  • [Journal Article] Design and Modelling of Cloud-based Burst Buffers (poster)2015

    • Author(s)
      Tianqi Xu, Kento Sato, Satoshi Matsuoka
    • Journal Title

      Supercomputing2015 (SC15)

      Volume: - Pages: -

    • Peer Reviewed / Open Access / Int'l Joint Research
  • [Journal Article] 多段階ブロッキングによるメモリアクセス量削減を図ったGPU向け疎行列ベクトル積計算手法の性能評価 (ポスター)2015

    • Author(s)
      長坂侑亮, 額田彰, 松岡聡
    • Journal Title

      GPU Technolgy Conference Japan (GTC Japan)

      Volume: - Pages: -

    • Peer Reviewed / Open Access
  • [Journal Article] mrCUDA: Low-Overhead Middleware for Live Migrating Remote GPU Execution to Local GPU Execution (poster)2015

    • Author(s)
      Pak Markthub, Akihiro Nomura, Satoshi Matsuoka
    • Journal Title

      GPU Technolgy Conference Japan (GTC Japan)

      Volume: - Pages: -

    • Peer Reviewed / Open Access
  • [Presentation] The Inevitable End of Moore’s Law beyond Exascale will Result in Data and HPC Convergence and More (Keynote Talk)2016

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      International Symposium on Grids and Clouds 2016
    • Place of Presentation
      Academia Sinica, Taipei Taiwan
    • Year and Date
      2016-03-15 – 2016-03-15
    • Int'l Joint Research / Invited
  • [Presentation] ビッグデータ・機械学習とスーパーコンピューティングの融合、およびその未来2016

    • Author(s)
      松岡 聡
    • Organizer
      Creater’s Talk
    • Place of Presentation
      デンソー東京支社(東京日本橋タワー)
    • Year and Date
      2016-03-09 – 2016-03-09
    • Invited
  • [Presentation] Collaborative Research between DoE Labs and Tokyo Tech GSIC on Extreme Scale Computing - Success Stories2016

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      The 6th AICS International Symposium
    • Place of Presentation
      RIKEN AICS, Kobe
    • Year and Date
      2016-02-22 – 2016-02-22
    • Int'l Joint Research / Invited
  • [Presentation] The Inevitable End of Moores Law beyond Exascale will Result in Disruptive Change (Plenary Talk)2016

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      2016 Conference on Advanced Topics and Auto Tuning in High-Performance Scientific Computing (ATAT in HPSC 2016)
    • Place of Presentation
      Mathematics Research Center, National Taiwan University, Taiwan
    • Year and Date
      2016-02-19 – 2016-02-19
    • Int'l Joint Research / Invited
  • [Presentation] FGPA Experience at Titech2016

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      ANL Workshop on Reconfigurable Computing for Post Moore Era
    • Place of Presentation
      Argonne National Laboratory, IL USA
    • Year and Date
      2016-01-21 – 2016-01-21
    • Int'l Joint Research / Invited
  • [Presentation] ポストムーア時代に向けたBYTES中心の高性能計算2015

    • Author(s)
      松岡 聡
    • Organizer
      光通信システムシンポジウムワークショップ
    • Place of Presentation
      東レ総合研修センター (静岡県三島市)
    • Year and Date
      2015-12-17 – 2015-12-17
    • Invited
  • [Presentation] Panel: Procuring Supercomputers: Best Practices and Lessons Learned (Panel Talk)2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      Supercomputong2015 (SC15)
    • Place of Presentation
      Austin Convention Center, TX USA
    • Year and Date
      2015-11-19 – 2015-11-19
    • Int'l Joint Research / Invited
  • [Presentation] Panel: Programming Models for Parallel Architectures and Requirements for Pre-Exascale (Panel Talk)2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      Supercomputong2015 (SC15)
    • Place of Presentation
      Austin Convention Center, TX USA
    • Year and Date
      2015-11-18 – 2015-11-18
    • Int'l Joint Research / Invited
  • [Presentation] Towards TSUBAME3 and beyond towards Inevitable End of Moore's Law (Booth Talk @NEC Booth)2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      Supercomputong2015 (SC15)
    • Place of Presentation
      Austin Convention Center, TX USA
    • Year and Date
      2015-11-18 – 2015-11-18
    • Int'l Joint Research / Invited
  • [Presentation] Inevitable Convergence of Big Data and HPC (Booth Talk @Tokyo Tech Booth)2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      Supercomputong2015 (SC15)
    • Place of Presentation
      Austin Convention Center, TX USA
    • Year and Date
      2015-11-17 – 2015-11-17
    • Int'l Joint Research / Invited
  • [Presentation] Japanese Supercomputer centers2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      6th Annual workshop Energy Efficient HPC Working Group Annual Workshop (EE HPC WG)
    • Place of Presentation
      Austin Convention Center, TX USA
    • Year and Date
      2015-11-16 – 2015-11-16
    • Int'l Joint Research / Invited
  • [Presentation] The Inevitable End of Moore’s Law beyond Exascale will Result in Data and HPC Convergence2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      HPC China2015
    • Place of Presentation
      Crowne Plaza Hotel Wuxi Taihu, China
    • Year and Date
      2015-11-11 – 2015-11-11
    • Int'l Joint Research / Invited
  • [Presentation] Convergence of extreme big data and extreme computing towards exascale and beyond2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      International Workshop on CO-DESIGN
    • Place of Presentation
      Crowne Plaza Hotel Wuxi Taihu, China
    • Year and Date
      2015-11-10 – 2015-11-10
    • Int'l Joint Research / Invited
  • [Presentation] The Inevitable End of Moore’s Law beyond Exascale will cause Big Data and HPC Convergence (Keynote Talk)2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      SBAC-PAD 2015
    • Place of Presentation
      Majestic Palace Hotel, Florianopolis, Brazil
    • Year and Date
      2015-10-20 – 2015-10-20
    • Int'l Joint Research / Invited
  • [Presentation] Directives and Tools for Accelerators: A Seismic Programming Shift (Keynote Talk)2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      Directives and Tools for Accelerators Workshop
    • Place of Presentation
      University of Houston Alumni Center, TX USA
    • Year and Date
      2015-10-13 – 2015-10-13
    • Int'l Joint Research / Invited
  • [Presentation] The Inevitable End of Moore’s Law beyond Exascale will Result in Data and HPC Convergence2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      Russian Supercomputing Days
    • Place of Presentation
      The Holiday Inn Moscow Sokolniki Hotel, Moscow, Russia
    • Year and Date
      2015-09-28 – 2015-09-28
    • Int'l Joint Research / Invited
  • [Presentation] The Adventures of Resource Scheduling with PBS Professional in a Heavily Accelerated, Energy Efficient TSUBAME2 Supercomputer for Everyone (Keynote Talk)2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      2015 PBS Works User Meeting
    • Place of Presentation
      Computer History Museum, CA USA
    • Year and Date
      2015-09-15 – 2015-09-15
    • Int'l Joint Research / Invited
  • [Presentation] Moderated Panel: Strategy to Deliver Exascale Science (Panel Talk)2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      2015 Smoky Mountains Computational Sciences and Engineering Conference (SMC2015)
    • Place of Presentation
      Park Vista Hotel, Gatlinburg TN USA
    • Year and Date
      2015-09-02 – 2015-09-02
    • Int'l Joint Research / Invited
  • [Presentation] The Inevitable End of Moore's Law beyond Exascale will Result in Data and HPC Convergence2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      The Information Science and Technology Seminar Series
    • Place of Presentation
      The Information Science & Technology Institute (ISTI) , Los Alamos National Laboratory, NM USA
    • Year and Date
      2015-08-26 – 2015-08-26
    • Int'l Joint Research / Invited
  • [Presentation] Current trends for high performance computing and challenges for the future2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      CIMI Semester2015
    • Place of Presentation
      INPT-ENSEEIHT, Camichel France
    • Year and Date
      2015-07-02 – 2015-07-02
    • Int'l Joint Research / Invited
  • [Presentation] Harnessing the deep memory hierarchy with communication reducing algorithms for Exascale2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      CIMI Semester2015
    • Place of Presentation
      INPT-ENSEEIHT, Camichel France
    • Year and Date
      2015-07-01 – 2015-07-01
    • Int'l Joint Research / Invited
  • [Presentation] Toward Inevitable Convergence of HPC and Big Data2015

    • Author(s)
      Satoshi Matsuoka
    • Organizer
      The Third Annual NCSA Blue Waters Symposium
    • Place of Presentation
      The Sunriver Resort Oregon, USA
    • Year and Date
      2015-05-13 – 2015-05-13
    • Int'l Joint Research / Invited
  • [Book] 「人工知能テクノロジー総覧」 第6章クラウドサーバーの進化 “ムーアの法則が終わった後のスーパーコンピューターの姿”(p22)2015

    • Author(s)
      松岡 聡ほか
    • Total Pages
      240
    • Publisher
      日経BP社

URL: 

Published: 2017-01-06   Modified: 2022-01-28  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi