1 / 6

NCSA RP Update: May 2010

NCSA RP Update: May 2010. John Towns Director, Persistent Infrastructure jtowns@ncsa.illinois.edu. In Brief… . Operationally continuing along as normal Retired Mercury March 31, 2010 access to files for 1 month to allow moving data 6 years of operation! Jan 2004 to Jan 2010

Download Presentation

NCSA RP Update: May 2010

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. NCSA RP Update:May 2010 John Towns Director, Persistent Infrastructure jtowns@ncsa.illinois.edu

  2. In Brief… • Operationally continuing along as normal • Retired Mercury March 31, 2010 • access to files for 1 month to allow moving data • 6 years of operation! • Jan 2004 to Jan 2010 • the last of the original DTF resources that started TeraGrid • Of note: Ember to arrive over next several months Imaginations unbound

  3. The Newest Addition: Ember • In summary: SGI Altix UV 1000 • 16.4 TF peak • 1536 Nehalem-EX cores • 8 TB memory • 300+ TB storage • Altix UV 10 front-end system • 24 cores • 4 x 6-core Nehalem-EX • 128 GB memory • Consists of four 4 SMPs • each SMP • 32 compute blades • 384 cores • 2 TB memory • each compute blade • dual Nehalem-EX (6 core) processors • 64 GB memory • 5.33 GB/core • NUMALink 5 • paired node 2D torus Imaginations unbound

  4. Dual Nehalem-EX processors Intel X7542 (Beckton) 6 cores/socket 18 MB cache 2.67 GHz QPI @ 5.86GT/s (23.2GB/s) 64 GB memory 16 x 4GB memory modules 34.1 GB/s per socket 5.68 GB/s per core Boxboro IOH I/O Riser QPI QPI Nehalem-EX Nehalem-EX BoxboroIOH QPI UV HUB (8) DDR3 RDIMMs & (4) Millbrook Memory Buffers per socket QPI QPI RLDRAM (Snoop Acceleration) (2) Directory FB-DIMMs (4) NUMAlink 5 Ember Compute Blade • UV Hub • acts as node controller • support for globally addressable memory • MPI Offload Engine • NUMALink 5 • 15 GB/s bi-directional • ~1µs latency

  5. Ember Environment • Storage • 300+ TB filesystem; 8+ GB/s • 24 SGI TP9500 disk systems • 146 GB Fiberchannel drives NOTE: currently negotiating a more favorable storage sub-system • Network connectivity • 10 x 10GigE connections to NCSA’s core network • 2 bonded per machine • three OC-192 circuits to Chicago (TeraGrid) • 10 GigE links to NLR PacketNet and the MREN GigaPOP • access to Internet2 and other research and education networks • Archive: 5 PB capacity • EMC/Legato DiskXtender (UniTree) • distributed across two SGI Altix 4700 servers • two ADIC S10K libraries • 10 x 10GigE connections to NCSA’s core network Imaginations unbound

  6. Deployment Schedule • Purchase approvals complete: March 10, 2010 • NSF funding secured • Univ of Illinois Board of Trustees Approval • March 10, 2010 meeting • Purchase Order issued: March 26, 2010 • Login node (UV 10) ship: May 14, 2010 • system build, OS, SGI environment • NCSA environment, user tools • internal testing by systems staff and user/applications support staff • First 384 core UV 1000 SMPs: mid July 2010 • applications testing and benchmarking by staff • early Friendly User access • Three additional 384 core UV SMPs: early August 2010 • further applications testing and benchmarking • full Friendly User access period • Full production: no later than Oct 1, 2010 Imaginations unbound

More Related