230 likes | 391 Views
KEK High Energy Accelerator Research Organization. HEP Data GRID in Japan. Yoshiyuki Watase Computing Research Center KEK. HEP Data Grid Workshop Nov. 8-9 2002 at CHEP KNU Korea. Outline. Network Infrastructure HEP Grid Activities in Japan ATLAS Belle R&D of GFarm
E N D
KEK High Energy Accelerator Research Organization HEP Data GRID in Japan Yoshiyuki Watase Computing Research Center KEK HEP Data Grid Workshop Nov. 8-9 2002 at CHEP KNU Korea
Outline • Network Infrastructure • HEP Grid Activities in Japan • ATLAS • Belle • R&D of GFarm • Related “Virtual Laboratory” activities • Collaboratory • GAN: Global Accelerator Network • Funding status for Grid • Conclusion HEP Data Grid Workshop @ KNU
Network Infrastructure 1 • SuperSINET (Science Network) • Jan. 2002, fiber optic network was provided by NII for research in Univ.’s and Institutes ( ~ up to 40 nodes). • 10 Gbps IP backbone and many p-to-p GbE links for research groups: HEP, Astronomy, Bioinformatics, Nano-tech, GRID computing • HEP • MPLS-VPN of HEPnet-J for ~25 university groups by sharing the 10 Gbps IP backbone • GbE links from KEK to Univ. ‘sTohoku, Tokyo, Nagoya, Kyoto, Osaka, Tsukuba, Titech(Tokyo Inst. of Tech), ICRR(Inst. Cosmic Ray Research) NII: National Institute for Informatics HEP Data Grid Workshop @ KNU
Tohoku U OXC NII Chiba IP 10 Gbps WDM path NIFS IMS IP router Nagoya U NIG Nagoya Hub Hokkaido Osaka Hub Tokyo Hub KEK Osaka U Tsukuba US,EU Kyoto U Korea NII KyushuU ISAS U Tokyo ICR Kyoto-U Internet TITEC NAO Waseda IMS U-Tokyo Network Topology of SuperSINET HEP Data Grid Workshop @ KNU
Network Infrastructure 2 • East bound: to US, EU • NII will implements 2.4 G x 2 terminated at NY on Jan. 2003. • We need transit line between NY and StarLight for connection to CERN with end-to-end GbE. • We need peering to ESnet at NY. • West bound: to Korea • Genkai project will provide ~ GbE connection between CHEP(KNU) and KEK. • Being demonstrated at APAN Fukuoka Meeting Jan. 22, 2003 HEP Data Grid Workshop @ KNU
Network Infrastructure TEIN NII- NY TRANSPAC Taiwan-US Japan (NII) -NY : 2.4G x 2 Jan. 2003 Japan – US: 622M x 2 (TRANSPAC) Korea – US: 45 M Korea – Japan: 2.4G Jan. 2003 Korea – Europe: 2 M (TEIN) China(IHEP) – Japan(KEK): 128 kbps (HEP) China – US: 10 M Taiwan – Japan: 155 M Taiwan – US: 622 M (Dec. 2002) HEP Data Grid Workshop @ KNU
LHC/Atlas-1 • Atlas Regional Center • Being set up at ICEPP (Int’l Center for Particle Physics, U. Tokyo) • Installation of PC farms and Storage • PC farm: 78 x P-III 1.4GHz + New PC farm :214 x Xeon 2.8 GHz (early 2003) • Tape TLO 7TB ( CASTOR) + 20-30 TB Disk data server (early 2003) • Data Challenge DC1 • Phase-1 (~ Sep. ): Monte Carlo production 5x107 events • Phase-2 ( Nov. ~ ) : Generate events with event-overlapping HEP Data Grid Workshop @ KNU
LHC/Atlas -2 • Globus installed in PC’s at KEK and ICEPP • Testing Globus environments • Grid environments for LHC exp. : LCG – 1 • Start test on Nov. 2002 • Testing between ICEPP and KEK • Testing between ICEPP and other sites in EU/US ( 2003 ) • Remote data access via GbE • Atlas simulation data at ICEPP being stored in HPSS storage at KEK through GbE link • Test Resources: ICEPP PC farm 78 cpu’s KEK PC farm 100cpu’s HPSSS ~ 20 TB HEP Data Grid Workshop @ KNU
LHC/Atlas - 3 Data Transfer Throughput Test KEK ICEPP, Tokyo GbE (~60 km) IBM RS6000/SP HPSS PC Farm PC Farms Single stream netperf: ~70 MB/s ftp : ~ 30MB/s simulation data ~ 10 TB netperf test over 2 streams Data Transfer Throughput Test over GbE 1. netperf test 2. ftp 3. HPSS access over GbE Local access ~ 30 MB/sec WAN access ~ being measured HEP Data Grid Workshop @ KNU
KEKB/Belle • B Factory experiment • Luminosity reached to 8 x 1033 /cm2/s in Oct. 2002: updating the world record and still improving • Data storage rate = ~ 500 GB/day • Data analysis • Resources at KEK: • 1000 PC’s for production and simulation • Storage ~ 630 TB • Resources at Univ.’s • Nagoya Tokyo Tohoku • Smaller resources in the collaboration institutes compared to BaBar exp. • BaBar has partners in IN2P3, RAL. • Need more funding for universities • planning to extend their computing power in Data Grid paradigm HEP Data Grid Workshop @ KNU
KEKB/Belle: Data Moving for Data Analysis over SuperSINET GbE e+e- Bo Bo Neutrino Center Tohoku U. 400 GB/day ~45 Mbps ~ 100GB/day NFS Osaka U. KEK 170 GB/day 1TB/day ~100Mbps • U. Tokyo Nagoya U. HEP Data Grid Workshop @ KNU
R&D of Gfarm(Grid data farm) • Gfarm http://datafarm.apgrid.org/ • Development by AIST, Titech, KEK since 2000 • Architecture • PC farm with large local disk/node as a GRID fabric • Large data file is divided into fragments and stored in the disks by read-in • Data file integrity is managed by the Gfarm metadata DB • Data I/O by parallel file system • Affinity scheduling of process and storage for data residence • Service daemon process: gfsd is running at each node • Authentication by gfarm key and/or GSI AIST :Advanced Inst. for Science and Technology HEP Data Grid Workshop @ KNU
Gfarm Test • PC farm used: Athlon 1.2GHz 16nodes • Fast Ethernet • Data replication between two 8-nodes • 8 parallel copy of 1GB fragments : 8 GB data file % gfrep –H <nodes> gfarm:userdata.bin • Throughput 89.4 MB/s vs. 90.2 MB/s (sum of independent transfers) Overhead in access to the metadata ~ 1% Metadata DB Server Gfarm <nodes> Data file gfarm:userdata.bin HEP Data Grid Workshop @ KNU
NII Gfarm Demonstration at SC2002 Nov. 18 “Bandwidth Challenge” Data transfer, data replication over trans-oceanic network HEP Data Grid Workshop @ KNU
Related activity : Collaboratory • “Virtual Laboratory” for Material Science • Funded 2001-2005 • 5 laboratories: KEK, IMS, U. Tokyo, Tohoku U., Kyoto U. • Implementation • Remote control of experimental equipments X-ray diffractometers at Photon Factory of KEK • Data sharing • Simulation by supercomputers linked • IP video conferencing system with application sharing • Demonstrated the prototype system Oct. 23, 2002 • Remote control from IMS(Nagoya) of diffractometers at the beam line of PF/KEK. • H.323 video conferencing with 4 other institutesIMS: Institute for Molecular Science HEP Data Grid Workshop @ KNU
Related activity: GANGlobal Accelerator Network • Aimed to remote operation of large accelerator facility, such as future Linear Collider • It facilitates acc control and monitoring for remote operation shift • International workshop held 2001, 2002 • KEK Planning to implement at J-PARC( Japan Proton Accelerator Research Complex) • 50 GeV, 3 GeV proton accelerators at Tokai 50 km away from Tsukuba(KEK) • Construction 2001 – 2006 JAERI Tokai Campus KEK HEP Data Grid Workshop @ KNU
Funding status for Grid • Gigabit Network MEXT: Ministry of Edu, Sci, .. • MEXTInformatics Project • A05: Application Grid projects for science ~ US$ 6M / 5 y • Networking supercomputers: ITBL ~ US$ 105M / 5y • AIST GTRC(Grid Technology Research Center)~ US$ 140M • Other Funding from Institutional Resources • LHC/Atlas(ICEPP,KEK), Gfarm, Virtual Observatory, Ninf-G, BioGrid(Osaka U ), OBIGrid(Riken),……. • Projects proposed (2003 ~ 5 years) • National Research Grid Initiative(US$ 360M): focused in Bio & Nanotech • National ‘Business’ Grid Project(US$ 250M): ASP business HEP Data Grid Workshop @ KNU
Gigabit network • Super SINET • 10Gbps MEXT DWDM nationwide network • About 40 universities and labs • US$ 60M/yr • 5 application areas: HEP, Nanotech, Bioinformatics, Astronomy, Grid Computing • Tsukuba-WAN • 330Gbps DWDM ring in the Tsukuba area • Tens of national labs • 6 supercomputers in TOP100 • Testbed for Grid infrastructure and applications HEP Data Grid Workshop @ KNU
MEXT Grid Projects • MEXT “Informatics” Project A05 “Grid” Area (US$ 6M, 5 years) • Lead by Shinji Shimozo(Osaka U.) & Satoshi Matsuoka(Titech) • 11 groups of universities and labs • Research and development on application, middleware, and infrastructure • Gfarm project is partly supported. • Tokyo Institute of Technology “Titech Campus Grid” (US$ 2M) • Lead by Satoshi Matsuoka(Titech) • Total of 800 CPU PC clusters at 13 locations in a 30km range connected by Super TITANET(1-4Gbps) backbone • Pilot project for virtual parallel computer infrastructure for high end computational e-Science HEP Data Grid Workshop @ KNU
ITBL • ITBL(IT-based Laboratory) • Government Labs: NAL, RIKEN, NIED, NIMS, JST, JAERI • Project period: 2001-2005 (3-stage project) with total of US$ 105M funding • Applications: mechanical simulation, computational biology, material science, environment, earthquake engineering • Step 1: Supercomputer centers of government lab arenetworked via SuperSINET • Step 2: “Virtual Research Environment”: Grid-enabling laboratory applications • Step 3: Sharing information among researchers from widely distributed disciplines and institutions HEP Data Grid Workshop @ KNU
Environmental Circulation Simulation for Pollutant Materials VPP300 (Vector Parallel Computer) Atmospheric Environment Simulation Two-Dimensional Data at Sea Surface Two-DimensionalData at Ground Surface Stampi Stampi COMPAQ α (High-Performance PC) Marine Environment Simulation AP3000 (Scalar Parallel Computer) Terrestrial Environment Simulation Large-scale Hartree-Fock Calculation SPring8 Fluid-Particle Hybrid Simulation for Tokamak Plasmas Control Diagonalization Orthonormalizarion Pool of task distribution Integral handling Partial accumulation Fij<-Fij+Dkl*qijkl Vector Machine Scalar Machine Electronic fluid /Electro-Magnetic field Ion Particles Stampi Vector Machine Scalar Machine Grid Computing Applications implementing key technologies of ITBL HEP Data Grid Workshop @ KNU
Grid Technology Research Center • Part of AIST under METI (Minitry of Economy, Trade and Industry) • GTRC(Grid Technology Research Center) • Established in Jan. 2002 • Director: Satoshi Sekiguchi • US$140M for building, equipment, R&D and operations • To be central institute for Grid research in Japan • Not just R&D, but serve as central CA, ApGrid’s principal site • Collaboration with institutions and other Grid projects of both domestic and international scale • Organizing GRID Consortium Japan HEP Data Grid Workshop @ KNU
Conclusions • Grid activities are starting. But not coordinated support for middleware development • International testbed for LHC/Atlas in 2003 • Possible collaboration in Asian region for KEKB/Belle • Network is emerging for heavy application users. • domestic and international use • Testbed of the Grid middleware in collaboration with CS people HEP Data Grid Workshop @ KNU