The Status Report on CDF Data Processing KPS, April 23-24, 2009 Daejeon Convention Center, Daejeon, Korea The Status Report on CDF Data Processing 정민호, 조기현*, 김현우 (KISTI 고에너지물리연구팀), 김동희, 양유철, 서준석, 공대정, 김지은, 장성현, 칸 아딜 (경북대 물리 및 에너지학부), 김수봉, 이재승, 이영장, 문창성, 정지은 (서울대, 물리학부), 유인태, 임규빈 (성균관대 물리학과), 주경광 (전남대 물리학과), 김현수 (전북대 물리학과) 오영도 (포항공대 물리학과), 전은주 (세종대 물리학과) * Corresponding author December 1, 2018

2018년 12월 1일

Ref. Dr. Donatella Lucchesi

CDF Grid (2009.4] Grid CAF Head node Work node Grid Middleware Method Data Handling NAM CAF Fermilab (USA) USSD (USA), etc. OSG Condor over Globus CDF Grid Fermilab (USA) Fermilab (USA), etc. Condor over Globs Lcg CAF CNAF (Italia) IN2P3 (France), etc. gLite WMS (Workload Management System) CNAF CAF (Iltaia) gLIte GlideinWMS Partially SAM+SRM Pac CAF AS (Taiwan) KISTI (Korea) Etc. OSG, gLite 4 4

Grid Site (KR-KISTI-HEP) Grid Farm (Pacific CAF) Pac CAF at KISTI Service Component Legend Repository Grid CAF Boundary Grid Site Boundary Grid Pool Job Data Transportation Control Data Grid Site (KR-KISTI-HEP) Worker Node (hep015) … (hep004) Work Node (hep003) dCache Pool (se-hep) Grid Farm (Pacific CAF) VOMS Authentication Glideins dCache User Interface CafGui CafSubmit CafMon CAF Head Node Submitter Monitor Collector Negotiator Schedd Globus .... Job Submission Web monitoring Job Output Condor CE (hep001) 25 core processors 30 TB 5 5

Monitoring of Pac CAF 6 6

Networks Mathis Equation BW: Upper bound of TCP throughput KEK GLORIAD-KR/10Gbps Hong Kong CSTNet-APAN-JP/1Gbps Fermilab Seattle ESnet/10Gbps CERN/ IN2P3 GLORIAD/1Gbps Chicago StarLight/10Gbps KISTI Mathis Equation KISTI is connected with KEK by 1 Gbps via Hongkong. KISTI is connected with Fermilab via 10 Gbps. KISTI is connected is CERN via 1 Gbps via Chicago. The bottle neck is inside of USA between Seattle and Chicago. Mathis Equation shows Bandwidth depends on Round Trip time and Packet Loss. Therefore, we study the packet loss and Round trip time. BW: Upper bound of TCP throughput MSS: Max Segment Size RTT: Round Trip Time P: Packet Loss

KISTI-Fermilab Jan. 6~20, 2009 (2 weeks) Packet loss is zero. RTT is stable. We avoid Chicago area congestions by direct network from Seattle. 2018년 12월 1일

CDF Resource Usage - Total Ref. Dr. Donatella Lucchesi 2018년 12월 1일

Offsite Usage – USA & Asia => CGCC (CDF Grid Computing Center) Donatella clearly mentioned about KISTI in International Finance Committee Meeting on November ’08.

CDF Grid Computing Center @ KISTI PacCAF vs. CGCC Pacific CAF @ KISTI CGCC @ KISTI Head node paccaf.phys.sinica.edu.tw New Head node VOMS server voms.cnaf.infn.it same Work nodes 25 core processes (gLite) 500 core processes (gLite, OSG) Storage 30 T byte 250 T byte Data Handling N/A Available 11

대용량 실험데이터 센터 (2009.4] - CDF, Belle, ALICE, DMRC, … CPU Storage 현재: 240 core process (run) 추가: 300 core process (5월말) Storage 현재: 50 TB (run) 추가: 200 TB (6월말) 계획: 그리드 연동 (12월말) Experiments CDF: 100 core + α (plan) ⇒ CGCC ALICE: 128 core (run) => ALICE Tier2 Belle: 64 core + batch (run) DMRC: 32 core + batch (run)

Summary Data processing for CDF has been developed from CAF (Central Analysis Farm) to Grid. KISTI is one of leading institutes of CDF data processing.

Thank you. (http://hep.kisti.re.kr)