Reprints from my posting to SAN-Tech Mailing List and ...

2011/06/21

[san-tech][02877] US NCSA Blue Waters I/O Hub and Interconnection Network情報とか

Date: Wed, 29 Dec 2010 20:14:59 +0900
------------------------------------------------
2011/06/23
[san-tech][01653] RAIT: Redundant Array of Independent Tape
------------------------------------------------
HPC寄りの話題ですが、先のメール

[san-tech][02875] Re[2]: NVIDIA GPUDirect等々 (+Storage と GPU)

で紹介した

> The fourth workshop of the Joint Laboratory for Petascale Computing
>  November 22-24, 2010
>   http://jointlab.ncsa.illinois.edu/events/workshop4/
> Workshop Program
>   https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program

を眺めていたら、いろいろ面白い資料がありました。

----------------------------------------
Blue Waters関連:
Blue Waters project
  http://www.ncsa.illinois.edu/BlueWaters/

"Blue Waters: A Super-System to Explore the Expanse and Depth of 21st
 Century Science"
 Bill Kramer, NCSA, USA
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-KramerA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-bkramer2.pdf

※設備構築パートナーに Yahoo!が含まれています
※Blue Waters Super-System (23枚目)
Blue Waters本体、ストレージ、ネットワークバンド幅が紹介
ストレージ:Sustained 500PB nea-line, 100GBps I/O
※34枚目以降に I/O Hub and Interconnection Network (Backupスライド)
HUB Chip: 1.12TB/s interconnection bandwidth
Collectives Acceleration Unit (CAU) 搭載
Operations
Reduce: NOP, SUM, MIN, MAX, OR, AND, XOR
Multicast
Integrated Switch Router (ISR)
3.0 GHz internal 56x56 crossbar switch

"Comparing archival policies for BlueWaters"
 Mathias Jacquelin, INRIA/ENS Lyon
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-JacquelinA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-mjacquelin.pdf

  "In this work, we introduce two archival policies tailored for the
   tape storage system that will be available on BlueWaters. We also
   show how to adapt the well known RAIT strategy (the counterpart of
   RAID policy for tapes) for BlueWaters."
※Tape Related Hardware
Tapes [約 500000本]: Each tape stores up to 1TB of uncompressed data
Tape Drives [約 500台]
Tape Libraries [約 3台]
Mover Nodes [約 50台]
※RAIT: Redundant Array of Independent Tapes


----------------------------------------
Exascale computing関係 (一部)

"Challenges on Programming Models and Languages for Post-Petascale
 Computing -- from Japanese NGS project "The K computer" to Exascale
 computing -- "
 Mitsuhisa Sato, U. Tsukuba, Japan
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-SatoA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-msato.pdf

"Toward Exascale"
 Marc Snir, UIUC, USA
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-SnirA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-msnir.pdf

"The UHPC X-Caliber Project"
 Arun Rodrigues, Sandia, USA
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-RodriguesA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-arodrigues.pdf
[san-tech][02527] Re: US DOE DARPA 新 HPCプロジェクト (OHPC) 公募 で
紹介した DARPA UHPC (Ubiquitous High Performance Computing) プロジェクト
Sandiaチーム (すごいメンバー)

[san-tech][02873] NVIDIA GPUDirect等々 (+Storage と GPU) で紹介した
> "GPU Computing: To ExaScale and Beyond"
>  Bill Dally, NVIDIA Research, November 18, 2010
>   http://www.nvidia.com/content/PDF/sc_2010/theater/Dally_SC10.pdf
も、UHPCプロジェクトの 1チームです (重なっているメーカもあります)。


----------------------------------------
で、ExaScale規模を考えたら安定性や CheckPointもすごく重要ということで、

"Framework for Event Log Analysis in HPC"
 Ana Gainaru, NCSA, USA
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-GainaruA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-againaru.pdf

"Clustering Message Passing Applications to Enhance Fault Tolerance
 Protocols"
 Esteban Menese, UIUC, USA
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-MenesesA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-emenese.pdf

"Latest Progresses on Rollback-Recovery Protocols for Send-Deterministic
 Applications"
 Thomas Ropars, INRIA, France
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-RoparsA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-tropars.pdf

"Transparent low-overhead checkpoint for GPU-accelerated clusters"
 Leonardo Bautista, Titech, Japan
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-GomezA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-lbautista.pdf
[san-tech][02875] Re[2]: NVIDIA GPUDirect等々 (+Storage と GPU) で
紹介した東工大 TSUBAME 2.0での研究

"On Scheduling Checkpoints of Exascale Application"
 Frederic Viven, INRIA/ENS Lyon, France
  https://wiki.ncsa.illinois.edu/display/jointlab/Workshop+Program#WorkshopProgram-VivienA
  https://wiki.ncsa.illinois.edu/download/attachments/17630761/INRIA-UIUC-WS4-fvivien.pdf

----------------------------------------
他にもアプリケーションや 3D FFTの最適化等々盛りだくさんです。

こちらも参考に
Joint Laboratory Publications
  http://jointlab.ncsa.illinois.edu/publications.html

0 件のコメント:

コメントを投稿