grid computing for high energy physics in japan hiroyuki matsunaga international center for...

26
Grid Computing for High Grid Computing for High Energy Physics in Japan Energy Physics in Japan Hiroyuki Matsunaga Hiroyuki Matsunaga International Center for Elementary Particle International Center for Elementary Particle Physics (ICEPP), Physics (ICEPP), The University of Tokyo The University of Tokyo International Workshop on e-Science for Physics International Workshop on e-Science for Physics 2008 2008

Upload: giles-norman

Post on 25-Dec-2015

214 views

Category:

Documents


0 download

TRANSCRIPT

Grid Computing for High Energy Grid Computing for High Energy Physics in JapanPhysics in Japan

Grid Computing for High Energy Grid Computing for High Energy Physics in JapanPhysics in Japan

Hiroyuki MatsunagaHiroyuki MatsunagaInternational Center for Elementary Particle Physics (ICEPP), International Center for Elementary Particle Physics (ICEPP),

The University of TokyoThe University of Tokyo

International Workshop on e-Science for Physics 2008International Workshop on e-Science for Physics 2008

2

Major High Energy Physics Program in JapanMajor High Energy Physics Program in JapanMajor High Energy Physics Program in JapanMajor High Energy Physics Program in Japan

KEK-B (Tsukuba)– Belle

J-PARC (Tokai)– Japan Proton Accelerator Research Complex– Operation will start within this year – T2K (Tokai to Kamioka)

long baseline neutrino experiment Kamioka

– SuperKamiokande– KamLAND

International collaboration– CERN LHC (ATLAS, ALICE)– Fermilab Tevatron (CDF)– BNL RHIC (PHENIX)

3

Grid Related ActivitiesGrid Related ActivitiesGrid Related ActivitiesGrid Related Activities

ICEPP, University of Tokyo– WLCG Tier2 site for ATLAS

Regional Center for ATLAS-Japan group Hiroshima University

– WLCG Tier2 site for ALICE KEK

– Two EGEE production sitesBELLE experiment, J-PARC, ILC…

– University support– NAREGI

Grid deployment at universities– Nagoya U. (Belle), Tsukuba U. (CDF)…

Network

4

Grid Deployment at University of TokyoGrid Deployment at University of TokyoGrid Deployment at University of TokyoGrid Deployment at University of Tokyo

ICEPP, University of Tokyo– Involved in international HEP experiments since 1974

Operated pilot system since 2002 Current computer system started working last year

– TOKYO-LCG2. gLite3 installed CC-IN2P3 (Lyon, France) is the associated Tier 1 site within A

TLAS computing model– Detector data from CERN go through CC-IN2P3– Exceptionally far distance for T1-T2

RTT ~280msec, ~10 hopsChallenge for efficient data transferData catalog for the files in Tokyo located at Lyon

– ASGC (Taiwan) could be additional associated Tier1Geographically nearest Tier 1 (RTT ~32msec)Operations have been supported by ASGC

– Neighboring timezone

5

Hardware resourcesHardware resourcesHardware resourcesHardware resources

Tier-2 site plus (non-grid) regional center facility

– Support local user analysis by ALTAS Japan group

Blade servers– 650 nodes (2600 cores)

Disk arrays– 140 Boxes (~6TB/box)– 4Gb Fibre-Channel

File servers– Attach 5 disk arrays– 10 GbE NIC

Tape robot (LTO3)– 8000 tapes, 32 drives

Pledged Planned to be pledged

2007 2008 2009 2010

CPU (kSI2k) 1000 1000 1000 3000

Disk (Tbyes) 200 400 400 600

Nominal WAN (Mbits/sec)

2000 2000 2000 2000

Tape robotBlade

serversDisk arraysDisk arrays

6

SINET3SINET3SINET3SINET3

SINET3 (Japanese NREN)– Third generation of SINET, since Apr. 2007– Provided by NII (National Institute of Informatics)

Backbone: up tp 40Gbps Major universities connect with 1-10 Gbps

– 10 Gbps to Tokyo RC International links

– 2 x 10 Gbps to US– 2 x 622 Mbps to Asia

7

International LinkInternational LinkInternational LinkInternational Link 10Gbps between Tokyo and CC-IN2P3

– SINET3 + GEANT + RENATER (French NREN)– public network (shared with other traffic)

1Gbps link to ASGC (to be upgraded to 2.4 Gbps)

TokyoNew York Lyon

SINET3 (10Gbps)GEANT (10Gbps)

RENATER (10Gbps)

Taipei

8

Network test with IperfNetwork test with IperfNetwork test with IperfNetwork test with Iperf

Memory-to-memory test performed with Iperf program Use Linux boxes dedicated for iperf test at both ends

– 1Gbps limited by NIC– Linux kernel 2.6.9 (BIC TCP)– Window size 8Mbytes, 8 parallel streams

For Lyon-Tokyo: long recovery time due to long RTT

Lyon <-> Tokyo (RTT: 280ms) Taipei <-> Tokyo (RTT: 32ms)

9

Data Transfer from Lyon Tier1 centerData Transfer from Lyon Tier1 centerData Transfer from Lyon Tier1 centerData Transfer from Lyon Tier1 center

Data transferred from Lyon to Tokyo– Used Storage Elements in production– ATLAS MC simulation data

Storage Elements– Lyon: dCache (>30 gridFTP servers, Solaris, ZFS)– Tokyo: DPM (6 gridFTP servers, Linux, XFS)

FTS (File Transfer System)– Main tool for bulk data transfer– Execute multiple file transfers (by using gridFTP) concurrent

lySet number of streams for gridFTP

– Used in ATLAS Distributed Data Management system

10

Performance of data transferPerformance of data transferPerformance of data transferPerformance of data transfer

>500 Mbytes/s observed in May, 2008– Filesize: 3.5Gbytes– 20 files in parallel, 10 streams each– ~40Mbytes/s for each file transfer

Low activity at CC-IN2P3 during the period (other than ours)

500 Mbytes/s

0 20 40Mbytes/s

10

1

100Throughput per file transfer

11

Data transfer between ASGC and TokyoData transfer between ASGC and TokyoData transfer between ASGC and TokyoData transfer between ASGC and Tokyo

Transferred 1000 files at a test (1Gbytes filesize) Tried various numbers of concurrent files / streams

– From 4/1 to 25/15 Saturate 1Gbps WAN bandwidth

Tokyo -> ASGC

ASGC -> Tokyo

20/10

4/2 4/4

16/1

20/1025/15

4/18/1 4/2 8/24/416/1

25/10 25/10

25/10

12

CPU Usage in the last year (Sep 2007 – Aug 2008)CPU Usage in the last year (Sep 2007 – Aug 2008)CPU Usage in the last year (Sep 2007 – Aug 2008)CPU Usage in the last year (Sep 2007 – Aug 2008)

3,253,321 CPU time (kSI2k*hours) in last year– Most jobs are ATLAS MC simulation

Job submission is coordinated by CC-IN2P3 (the associated Tier1)Outputs are uploaded to the data storage at CC-IN2P3

– Large contribution to the ATLAS MC production

Top 10 Tier2 site for ATLAS VO

0500000

100000015000002000000250000030000003500000400000045000005000000

AGLT

2

IN2P

3-CC

-T2

MWT2_U

CGRI

F

TOKY

O-LCG2 WT2

BU_A

TLAS

_Tier

2

MWT2_IU

UKI-N

ORTHG

RID-

MAN-H

EP

OU_OCH

EP_S

WT2

Nor

mal

ized

CP

U t

ime

atlas

TOKYO-LCG2 CPU time per month CPU time at Large Tier2 sitesTOKYO-LCG2 CPU time per month

0

100000

200000

300000

400000

500000

600000

7-Sep 7-Oct 7-Nov 7-Dec 8-Jan 8-Feb 8-Mar 8-Apr 8-May 8-Jun 8-Jul 8-Aug

Nor

mal

ized

CP

U t

im

atlas

13

ALICE Tier2 center at Hiroshima UniversityALICE Tier2 center at Hiroshima UniversityALICE Tier2 center at Hiroshima UniversityALICE Tier2 center at Hiroshima University

WLCG/EGEE site – “JP-HIROSHIMA-WLCG”

Possible Tier 2 site for ALICE

14

Status at HiroshimaStatus at HiroshimaStatus at HiroshimaStatus at Hiroshima

Just became EGEE production site– Aug. 2008

Associated Tier1 site will likely be CC-IN2P3– No ALICE Tier1 in Asia-Pacific region

Resources– 568 CPU cores

Dual-Core Xeon(3GHz) X 2cpus X 38boxesQuad-Core Xeon(2.6GHz) X 2cpus X 32boxesQuad-Core Xeon(3GHz) X 2cpus X 20blades

– Storage: ~200 TB next year

Network: 1Gbps– On SINET3

15

KEKKEKKEKKEK

Belle experiment has been running– Need to have access to existing peta-bytes of data

Site operations– KEK does not support any

LHC experiment– Try to gain experience by

operating sites in order to prepare for future Tier1 level Grid center

University support NAREGI

KEK Tsukuba campus

Mt. Tsukuba

KEKB

Linac

Belle exp.

16

Grid Deployment at KEKGrid Deployment at KEKGrid Deployment at KEKGrid Deployment at KEK

Two EGEE sites– JP-KEK-CRC-1

Rather experimental use and R&D– JP-KEK-CRC-2

More stable services NAREGI

– Used beta version for testing and evaluation Supported VOs

– belle (main target at present), ilc, calice, …– Not support LCG VOs

VOMS operation– belle (registered in CIC)– ppj (accelerator science in Japan), naokek– g4med, apdg, atlasj, ail

17

Belle VOBelle VOBelle VOBelle VO

Federation established– 5 countries, 7 institutes, 10 sites

Nagoya Univ., Univ. of Melbourne, ASGC, NCU, CYFRONET, Korea Univ., KEK

VOMS is provided by KEK Activities

– Submit MC production jobs– Functional and performance tests– Interface to existing peta-bytes of data

18

Takashi Sasaki (KEK)

19

ppj VOppj VOppj VOppj VO

Federated among major universities and KEK– Tohoku U. (ILC, KamLAND)– U. Tsukuba (CDF)– Nagoya U. (Belle, ATLAS)– Kobe U. (ILC, ATLAS)– Hiroshima IT (ATLAS, Computing Science)

Common VO for accelerator science in Japan– NOT depend on specific projects, but resources shared

KEK acts as GOC– Remote installation– Monitoring

Based on Nagios and Wiki– Software update

20

KEK Grid CAKEK Grid CAKEK Grid CAKEK Grid CA

Started since Jan. 2006 Accredited as an IGTF

(International Grid Trust Federation) compliant CA

JFY 2006

Apr 2006 - Mar 2007

JFY 2007

Apr2007 – Mar 2008

Personal cert. 68 119

Host cert. 139 238

Web server cert. 4 0

Numbers of Issued certificates

21

NAREGINAREGINAREGINAREGI

NAREGI: NAtional REsearch Grid Initiative– Host institute: National Institute of Infrmatics (NII) – R&D of the Grid middleware for research and indus

trial applications – Main targets are nanotechnology and biotechnolog

yMore focused on computing gridData grid part integrated later

Ver. 1.0 of middleware released in May, 2008– Software maintenance and user support services w

ill be continued

22

NAREGI at KEKNAREGI at KEKNAREGI at KEKNAREGI at KEK

NAREGI- version installed on the testbed– 1.0.1: Jun. 2006 – Nov. 2006

Manual installation for all the steps

– 1.0.2: Feb 2007– 2.0.0: Oct. 2007

apt-rpm installation

– 2.0.1: Dec. 2007

Site federation test– KEK-NAREGI/NII: Oct. 2007– KEK-National Astronomy Observatory (NAO): Mar. 2008

Evaluation of application environment of NAREGI– job submission/retrieval, remote data stage-in/out

23

Takashi Sasaki (KEK)

24

Data Storage: GfarmData Storage: GfarmData Storage: GfarmData Storage: Gfarm

Gfarm: distributed file system– DataGrid part in NAREGI– Data are stored in multiple disk servers

Tests performed :– Stage-in and stage-out to the Gfarm storage– GridFTP interface

Between gLite site and NAREGI site

– File access from applicationHave access with FUSE (Filesystem in userspace)

– Without the need of changing application program– IO speed is several times slower than local disk

25

Future Plan on NAREGI at KEKFuture Plan on NAREGI at KEKFuture Plan on NAREGI at KEKFuture Plan on NAREGI at KEK

Migration to the production version Test of interoperability with gLite Improve the middleware in the application dom

ain– Development of the new API to the application

Virtualization of the middlewarefor script languages (to be used at web portal as well)

– MonitoringJobs, sites,…

26

SummarySummarySummarySummary

WLCG– ATLAS Tier2 at Tokyo

Stable operation

– ALICE Tier2 at HiroshimaJust started operation in production

Coordinated effort lead by KEK– Site operations with gLite and NAREGI middlewares

Belle VO: SRB – Will be replaced with iRODs

ppj VO: deployment at universities– Supported and monitored by KEK

– NAREGIR&D, interoperability