International HEP Data Grid Workshop 9 Nov 2002

International HEP Data. Grid Workshop 9 Nov 2002 Application of the EDG Testbed Bockjoo Kim*, Soo-Bong Kim Seoul National University (SNU) Kihyeon Cho, Youngdo Oh, Dongchul Son Center for High Energy Physics (CHEP) Kyungpook National University Jysoo Lee KISTI, Supercomputing Center

Outline n LHC Computing Challenge n European Data. Grid (EDG) testbed in Korea n The EDG testbed and tests n The GDMP (Datamanagement) and tests n Possible Immediate Application of the EDG Testbed n Summary

LHC HEP computing, key parameters All LHC experiments at CERN: ¾ 10 Peta Bytes/yr data storage; disk: 2 P Byte Multi-experiment Tier 1: ¾ 3 Peta Byte/yr; disk: 0. 5 P Byte Tier 0 & 1 at CERN: 20 SI 95) 2 M SI 95 (PC today ~ Multi-experiment Tier 1: 0. 9 M SI 95 Networking Tier 0 --> Tier 1: 622 Mbps (4 Gbps) (black fibre: 1 Tbps today)

Online Data Rate vs Size Level 1 Rate (Hz) “How can this data reach the end user? ” It doesn’t… Factor O(1000) Online data reduction via trigger selection High No. Channels High Bandwidth (500 Gbit/s) 106 LHCB 105 104 HERA-B ATLAS CMS KLOE CDF II High Data Archive (Peta. Byte) CDF 103 H 1 ZEUS 102 104 UA 1 NA 49 105 LEP ALICE 106 107 Event Size (bytes)

European Data. Grid (EDG) To cope with the data challenges in LHC era Participants EC+(Korea+U. S. +Japan+China …) Operational at EC Specific app layer VO common application layer EDG middleware GLOBUS 2. 0 ALICE ATLAS CMS LHCb Other apps High level GRID middleware Basic Services (Globus) OS & Net services Concept of EDG WP 1: workload WP 2: data management WP 3: grid monitoring services WP 4: fabric management WP 5: mass storage management WP 6: Integration and testbed WP 7: network services WP 8: HEP application WP 9: Earth Observation WP 10: Biology WP 11: Dissemination WP 12: Management

The EDG Fabric and Testbed LCFG : EDG software + Profile + CE, WN, SE, RB, UI installation&conf UI : User Interface : Job submission interface for authorized users CE : Farms of commodity PCs § Globus Resource Access Management (GRAM) service+GSI RB : Resource Broker : Core of job scheduling SE : Any storage system: accessible via ftp, nfs, etc & GSI WN : Worker CPU hidden underneath CE ~ clusters An EDG Testbed CE/WN (PC Cluster) Listens update Profiles UI SE (GDMP) LCFG Server RB RPMs repository Profile repository All the machines are linux red hat

EDG Important Ingredients VO Management X. 509 Authentication + Users certs on LDAP Security Grid Information Index Service (GIIS): not fully used. RB does the job Resource Locating Virtual Organization(VO) : CMS, Atlas, etc Virtual Users and Groups for VO LDAP services for VO, authentication list with edg schema Replica Catalog (RC) uses also LDAP server with edg schema Replica Management uses GDMP (grid data mirroring pkg. ) Data Management

EDG Testbed in Korea UI Real user GDMP client (with new VO) 디스크 RB GSIFTP SE VOuser CE VOuser NFS GSIFTP 빅팻 디스크 Big Fat Disk NFS GDMP client (with new VO) NFS WN VOuser MAP on disk With maximum security GSIFTP In operation GDMP server (with new VO) grid-security SNU In operation KNU LDAP Server SNU In preparation SKKU. . . The EDG testbeds are installed at KNU and at SNU The globus simple CA is managed at KNU and at SNU to sign certs In addition to the default VO’s in EDG, a cdf VO is constructed Authentication and VO LDAP server is running at SNU A replica catalog (RC) LDAP server is running at SNU See the Korean EDG testbed description web site at http: //neutrino. snu. ac. kr/~bockjoo/EDG_testbed/ and EDG web

Authorization and grid-mapfile X. 509 certs by CA CA published on LDAP Check Auth. dir and VO o=xyz, dc=edg, dc=org ou=People CN=Mario Rossi CN=John Smith Authentication Certificate o=testbed, dc=edg, dc=org ou=tb 1 VO Directory CN=Franz Elmer Authentication Certificate ou=People ou=Admin CN=John Smith “Authorization Directory” Authentication Certificate mkgridmap local users CN=Franz Elmer grid-mapfile ban list

LDAP Services for Korean HEP DG Auth LDAP for RB job There also the CMS and the CDF RC More Info: http: //neutrino. snu. ac. kr/~bockjoo/EDG_testbed/ CDF VO LDAP CMS VO LDAP

How to Use the EDG Testbed Certificate should be obtained Account on a UI must exist UI Commands Basic Globus Commands grid-proxy-init To Access Resource Broker globus-url-copy ………………. EDG Commands(python) dg-job-cancel dg-job-get-output dg-job-get-logging-info dg-job-id-info dg-job-list-match dg-job-status dg-job-submit More Info: http: //marianne. in 2 p 3. fr/datagrid/documentation

An Illustration of Job Submission dg-job-cancel dg-job-get-output dg-job-get-logging-info dg-job-id-info UI dg-job-list-match dg-job-status JDL dg-job-submit Input “sandbox” LFN->PFN Output “sandbox” Job Query Job Submit Author. &Authen. Data Management Services Information Service Input “sandbox” Resource Broker Job Status Storage Element Brokerinfo Job Submission Service Logging & Book-keeping Job Status Output “sandbox” Compute Element

Job Submission - UI Job submission command Job Id

Job Submission - RB Initial job submission phase Job execution phase

Job Submission - JSS Server Log JSS parsed

Job Submission - Condor. G just located the resource (CE)

Job Submission – the CE=QUEUE

Job Submission – RB ftp+WN RB wuftpd WN pbs mom

GDMP: Datamanagement More Info: http: //project-gdmp. web. cern. ch/project-gdmp/

How to use GDMP and RC? 1. 2. Setup environment, e. g. , EDG_VO=cms Subscribe to other site’s SE gdmp_host_subscribe –S $SE 1 –P $P 1 –r $SE 2 –p $P 2 3. Create file(s) on SE by hand or by JDL (The file should belong tothe group $EDG_VO) 4. On UI, register the file locally gdmp_register_local_file -R -p /flatfiles/SE 00/cms/dummy –S $SE –P $P –V $EDG_VO 5. On UI, publish the file and register it on the RC gdmp_publish_catalogue -V cms -C -L log_ID 6. On UI, replicate file if necessary (gsiftp+ldapadd) gdmp_replicate_get –r $SE –p $P –V $EDG_VO More Info: http: //neutrino. snu. ac. kr/~bockjoo/EDG_testbed/contents/gdmp_test. html

Data Management using GDMP SE Subscription Check catalog on se 1 File Creation on SE LDAP : RC for se 1 Check File on SE Replicate from se 1 to se 2 Register and Publish LDAP: RC for se 1 and se 2 Check catalog on se 2

Example Application of EDG UI Real user GDMP client (with new VO) 디스크 RB GSIFTP Que CE VOuser SE VOuser NFS GSIFTP NFS Que CDF Software GSIFTP 빅팻 디스크 Big Fat Disk NFS GDMP client (with new VO) NFS WN VOuser GSIFTP MAP on disk With maximum security Que K 2 K Software GDMP server (with new VO) grid-security EDG testbed at SNU Currently running Experiment Existing Resources Matter of adding ques and hacking : examples exist Some works are in progress to include the existing node as a WN for different VO’s within a testbed site

Summary At the moment, the EDG testbed is operated at EC EDG testbeds are constructed and working at KNU and SNU in Korea For Data Management, an LDAP for RC is constructed and the GDMP is used within Korea. This is working. The testbed is also operational between KNU and SNU A possible HEP application for currently running experiments is illustrated and we are moving toward that direction The EDG testbed will be constructed for other institutions in Korea New VO will be tested extensively for a few months Hopefully, we would like to learn other testbeds and test crossgrid functionalities

Network to the world APII-Trans. Pac Europe CERN TEIN US FNAL 현해 한-일: 8 Mbps 한-미: 45 Mbps 10 Mbps 2 Gbps ( soon ) • 45 Mbps by the end of 2002

File transfer test on the network q parallel ftp used. Ø KNU – SNU : ~50 Mbps Ø KNU – SKKU : ~50 Mbps Ø SKKU – SNU : ~50 Mbps (155 Mbps) Ø CHEP – KEK : ~7 Mbps (8 Mbps) Ø CHEP – Fermilab : ~17 Mbps (45 Mbps) Ø CHEP – CERN : ~8 Mbps (10 Mbps) Ø FERMILAB – CERN : ~2 Gbps ( Data. TAG project of i. VDGL )

File transfer test on the network (KNU-SKKU, SNU) file size for one ftp= 30 MB

File transfer test on the network KNU-US ftp test US-CERN Data. TAG link test

Plans q Route to Europe Ø 1 st TEIN : 45 Mbps from next year Ø 2 nd : via USA Korea FERMILAB CERN % We will use Fermilab’s storage as buffer. Ø 3 rd : via Japan and USA Korea KEK Fermilab CERN 2 Gbps >1 Gbps 2 Gbps Participation to i. VDGL anables 2 nd and 3 rd route. q Contribution : 2003 CMS summer production We will provide significant CPU power and storage.

Future bandwidth requirement on the network q Korea – Japan – USA Connection Ø for experiments in Japan ~ 400 Mbps by 2006 ~ 650 Mbps after 2006 Ø CDF experiment ~40 Mbps by 2004 ~230 Mbps after 2004 Ø CMS experiment ~800 Mbps ( 1 PB/year ) è Including other traftic and Japan’s, at least 2 Gbps required. è Should evaluate to Lambda (10 Gbps) connection to USA, Japan, Europe

Future bandwidth requirement on the network 622 Mbps +10 Gbps l