1. A Design for KCAF for CDF Experiment Kihyeon Cho (CHEP, Kyungpook National University) and Jysoo Lee (KISTI, Supercomputing Center) The International Workshop on HEP Data Grid November 8-9, 2002

2. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 2  CHEP and Fermilab CDF  CDF Data/Analysis Flow  CDF Grid RoadMap (bottom –up method) □Step 1. CAF (Central Analysis Farm) □Step 2. DCAF (DeCentralized Analysis Farm) □Step 3. Grid  KCAF (DeCenteralized Analysis Farm in Korea)  Future Plan Contents

3. CHEP and Fermilab CDF Grid Korea CHEP Fermilab in USA KCAF (DCAF in Korea)

4. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 4 Server s … IBM 8271 PCs L3 Switch (3 군 ) Gigabit Ethernet C6509 KOREN Gigabit Switch (CHEP) … Servers Gigabit Ethernet FNAL KEK APII 45Mbps APII 2 X 1Gbps CERN TEIN 10Mbps Network between CHEP and overseas

5. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 5 Overview of Fermilab Main Injector and Recycler  p source Booster CDF D0 Fixed Target Experiment

6. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS l Totals ä 11 countries ä 52 institutions ä 525 physicists North AmericaEurope Asia 3 Natl. Labs 25 Universities 2 Universities 1 Research Lab 6 Universities 1 University 4 Universities 2 Research Labs 1 University 1 Research Lab 4 Universities 1 University KOREA Center for High Energy Physics: Kyungpook National University Seoul National University SungKyunKwan University CDF Collaboration

7. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 7 Production Farm

8. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 8 CDF Computer Resource CAF Stage 1 Current configuration: Stage 1 –63 dual worker nodes. –Roughly 160 GHz total, compared to 38 GHz for fcdfsgi2, and 8 GHz for cdfsga (Run I). –7 fileservers for physics (~50% full), 5 for DH, 2 for development, 2 for user scratch. FCC (Feynman Computer Center)

9. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 9 CAF design considerations: –Submit jobs from 'anywhere' –Job output can be: sent directly to desktop or stored on CAF for later retrieval or –input to subsequent job CAF Design Integer

10. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 10 Requirement in 2005 –200 simultaneous users will analyze 10 7 events of secondary data set in a day. –Need ~700 TB of disk and ~5THz of CPU by end of FY’05 Current CAF (Central Analysis Farm) is not enough: –Limited resources and spaces at FCC –In case of network problems at FCC, it is dangerous.  We need DCAF (DeCentralized Analysis Farm) and Grid DCAF (DeCentralized Analysis Farm) –Users around regional area and/or around the world –Korea, Toronto, Karlsruhe, …. –Let us call DCAF in Korea as “KCAF” DCAF/Grid at CDF Experiment

11. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 11 CAF in USA Center for HEP (CHEP) Kyungpook National U. Daegu, Korea KCAF (DeCentralized Analysis Farm in Korea) APII (Asia Pacific Information Infrastructure) DCAF in Karlsruhe DCAF in Toronto DCAF for CDF Experiment

12. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 12 Proposed Schemes for Final Goal DCAF in Korea (KCAF) CAF or

13. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 13 Step 1. To make MC production farm using KCAF –First, we start to construct 12 CPU test bed for KCAF. –After policy inside of CHEP (another test bed for EDG and iVDGL), we will decide how many CPUs for actual MC generation farm will be used among this year’s planed 140 CPUs. Step 2. To handle real data –To extend the KCAF to the real data handling system using SAM (Sequential data Access via Meta-data), Gridftp, etc after settling down real data handling system. Step 3. Final goal of CDF Grid –A gridification for KCAF related with EDG, iVDGL and CDF Grid The Road Map for Goal

14. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 14 1.Software CDF Software FBSNG for batch job 2.Kerborse Client : login from Korea to Fermilab Server : login from Fermilab to Korea KDC : Client & Server Trust 3.Data Handling System Glasgow, Italy and Yale use Data Handling System. SAM station needs to be installed on 1 PC work node. 4.OS for Fermilab Red Hat Linux 7.3 with Kernel 2.4.18 5.Fundamental license problem for DB The Road Map in Detail

15. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 15 1TB buffer (FCC) CAFGUI ICAFGUI SAM Station (KCAF) User Desktop Remote Desktop KCAF Head Node rcp, GridFTP bbftp cp KCAF Cluster FCC (CAF) CHEP (KCAF) ICAF FTP server CDFen (Raw Data) dCache rcp fcdfsam Smaster FSS, Stager stager Technical Request STKen Calibration Data dCache rcp A Design of KCAF for CDF (Fermilab) Grid

16. KIHYEON CHO CENTER FOR HIGH ENERGY PHYSICS 16 Around 140 CPU linux clusters will be constructed at the end of this year at CHEP. We are constructing KCAF for CDF at CHEP, Kyungpook National University. Contribute 1 T byte hard disk (now) + 4 T byte hard disk (December) to CDF for the network buffer between CHEP and Fermilab. The DCAF demonstration will be shown at SC2002 on November 16, 2002. Future Plan (This year)