180 likes | 311 Views
FZU Computing Centre. Jan Švec Institute of Physics of the AS CR, v.v.i. 29.8.2011. FZU Computing Centre. Computing center is an independent part of the Department of Networking and Computing Techniques.
E N D
FZU Computing Centre Jan Švec Institute of Physics of the AS CR, v.v.i. 29.8.2011
FZU Computing Centre • Computing center is an independent part of the Department of Networking and Computing Techniques. Members of the team lead by J. Chudoba from the Department of Experimental Particle Physics: T. Kouba, J. Švec, J. Uhlířová, M. Eliáš, J. Kundrát. • Operation is strongly supported by some members of the Department of detector development and data processing lead by M. Lokajíček: J. Horký, L. Fiala.
FZU Computing Centre • HEP experiments • D0 experiment, Fermilab (USA) • LHC experiments ATLAS, ALICE (CERN) • STAR experiment, BNL (USA) • Solid state physics • Astroparticle physics • Pierre Auger Observatory (Argentina)
History: 2002 • 34 HP LP1000 • 2x 1.3 GHz Pentium 3 • 1 GB RAM • First 1 TB of disks • “Terminal room”
History: 2004 • A real data center • 200 kVA UPS • 380 kVA diesel • 2x 56 kW CRACs • 67 HP DL140 • 3.06 GHz Prescotts • 10 TB disk storage
2007 18 HP BL460c 12 HP BL465c History: 2006 - 2007 • 2006 • 36 HP BL35p • 6 HP BL20p
History: 2006 – 2007 (cont.) • 3 HP DL360 for Xen • FC infrastructure • HP EVA • 70+ TB usable capacity, FATA drives • Disk images for Xen machines • Rest used as DPM • Warranty ended in Dec 2010 • A new box cheaper than warranty extension
History: 2008 • 84x IBM iDataPlex node dx340 • 2x Xeon E5440 => 8 cores • 20x Altix XE 310 twins (40 hosts) • 2x Xeon E5420 => 8 cores • 3x Overland Ultamus 4800 (48TB raw each) SAN • Tape library NEO 8000 • 2x VIA based NFS storage • First decommissioning of computing nodes • 2002's HP LP1000r
History: 2009 • 65x IBM iDataPlex dx360M2 nodes • 2x Xeon E5520 with HT => 16 Cores • 9x Altix XE 340 twins (18 hosts) • 2x Xeon E5520 without HT => 8 cores • All water cooled • 3x Nexsan SataBeast (84TB raw each) SAN • 3x Atom-based storage nodes • Wrong idea, WD15EADS-00P8B0 are just weird
History: 2009 • SGI Altix ICE 8200 • Solid state physics • 512 cores (128 x E5420 2.5G) • 1TB RAM • Infiniband • 6TB disk array (Infiniband) • Torque/Maui, OpenMPI
Cooling 2009 • 2009: New water cooling infrastructure • STULZ CLO 781A 2x 88 kW • Additional ceiling for hot-cold air separation
History: 2010 • 26x IBM iDataPlex dx360M3 nodes • 2x Xeon X5650 with HT => 24 Cores • Bull Optima 1500 – HP EVA replacement • 8x Supermicro SC847E16 + 847E16-RJBOD • DPM pool nodes • 1,3PB • Second decommissioning of computing nodes • 2004's HP DL140 • 8 IBM dx340 swapped for dx360M3
Network facilities • External connectivity delivered by CESNET, a GEANT stakeholder • 10Gb to public Internet • 1Gb to FZK (Karlsruhe) • 10Gb connection demuxed to several dedicated 1Gb lines • FNAL, BNL, ASGC, Czech Tier-3s,... • 10G internal network • Few machines still on 1G switches → LACP
2011: Current Procurements • Estimates: • Worker nodes, approx. 5500 HEPSPEC06 • 800TB disk storage • Water cooled rack + water cooled back doors • Another NFS fileserver
Current numbers • 275 WNs HEP, 76 WNs solid state physics • 3000 cores HEP, 560 cores solid state physics • Torque & Maui • 2PB on disk servers (DPM or NFSv3)
Questions? Q & A