ATLAS COMPUTING AT OU

Transcription

1 ATLAS COMPUTING AT OU Outline HORST SEVERINI OU DOE REVIEW FEBRUARY 1, 2010 Introduction US ATLAS Grid Computing and Open Science Grid (OSG) US ATLAS Tier 2 Center OU Resources and Network Summary and Outlook

2 Introduction OUHEP involved in computing efforts for both ATLAS and DØ Long standing involvements in ATLAS Data Challenges (DC) and DØ MC and Data Reprocessing efforts Also very active in various Open Science Grid (OSG) activities in recent years Working closely with Langston University and OSU as part of OCHEP Also with LU, UT Arlington, and UNM as part of the US ATLAS SW Tier 2 Center Using OUHEP desktop cluster, small OSG testbed cluster, medium OCHEP Tier 2 cluster, and large OSCER resources 2 OU DOE Review

3 US ATLAS Grid Computing and Open Science Grid OUHEP part of ATLAS Grid Computing and the OpenScienceGrid (OSG) since its very beginning OUHEP efforts include testing, integration, and deployment of OSG infrastructure for ATLAS computing: Early install and testing of development OSG releases Debugging and bug fixing Participate in integration efforts Installation and continuous monitoring of production releases Latest cycle of integration (1.1.7) and deployment (1.2.6) just finished 3 OU DOE Review

4 US ATLAS Grid Computing and Open Science Grid OU also involved in OSG Monitoring and Accounting (Karthik) RSV: Resource and Service Validation service; integration and debugging, as well as testing new features Gratia: Job Accounting and History; integration, debugging, and web interface updates GIP: Generic Information Provider; improving installed capacity reporting : dynamic automated reporting on site computing and storage capacities 4 OU DOE Review

5 US ATLAS Grid Computing and Open Science Grid Recent involvement in OSG Education and Outreach (Horst) Invited to co-lead several international OSG grid schools: Johannesburg, South Africa, and Sao Paulo, Brazil OSG Liaison for new GridUNESP VO; visit to Sao Paulo to help founding of OSG VO and site installation (DOSAR project) Valuable because of our cross disciplinary expertise in both OSG grid middleware and ATLAS software 5 OU DOE Review

6 US ATLAS Tier 2 Center US ATLAS SW Tier 2 Center (OU in Collaboration with LU, UTA, UNM) very successful Main focus is computing, in particular Grid computing and Distributed Analysis All of the above also major components of OCHEP efforts PanDA (Production and Distributed Analysis) Production (both MC and DA) running very successfully on OCHEP Tier 2 cluster Also involved in US ATLAS network throughput efforts: testing and improvement of PerfSonar (Karthik) Other OU Resources for PanDA Production are being worked on: large OSCER cluster (Sooner) and OU Condor pool (Horst) 6 OU DOE Review

7 US ATLAS Tier 2 Hardware 61 Node (260 Core) 2.33/3.2 GHz Xeon-64, 2 GB RAM per Core 10 Support Nodes (5 head, 5 storage) 2.33/3.2 GHz Xeon TB of usable DDN/IBRIX3 storage (24 TB raw) ROCKS 4.1 (RHEL4 64 bit) OSG tier2-01: head node tier2-02: storage transfer node (gsiftp) tier2-05: SRM (storage resource manager) Monitoring: Ganglia, MonALISA, Gratia, RSV, cron scripts 7 OU DOE Review

8 US ATLAS Tier 2 Hardware Just ordered additional 200 TB high performance (Lustre) storage and additional 34 dual-quad compute nodes This will greatly improve our ability for MC generation throughput and particularly data analysis (requires large amounts of disk space) But the increased cluster size will also require more man power; Problem: Karthik is funded by ending EPSCoR grant 8 OU DOE Review

9 US ATLAS Tier 2 Hardware 9 OU DOE Review

10 10 OU DOE Review

11 Other OUHEP Resources OUHEP Tier 3 cluster: 39 Node (77 CPU) 2 GHz P4/Xeon, 20 TB storage OSG Production site, OUHEP SAM station, OSG SAM station Used for DØ SAMGrid production, ATLAS MC, and local theory calculations About to add 4 more dual-quad nodes with additional 40 TB with $20k ARRA funds Will be used for ATLAS Analysis and PROOF farm OUHEP ITB: 8 Node 1.4 GHz P4, 80 GB storage OSG Integration site Used for OSG and SAMGrid integration testing 11 OU DOE Review

12 Other OU Resources Large OSCER MPI cluster, Sooner 534 Node (4272 core) 2.0 GHz Xeon-6, 150 TB storage Used for DØ computing as available (opportunistic) Will be used for ATLAS Tier 2 computing as soon as ATLAS DDM local site mover available 750 Node Condor pool: 3.0 GHz P4, 1 GB RAM, 40 GB HD, 100 Mbps network Distributed over Campus PC labs WinXP Host OS with CoLinux and Condor inside Used for DØ computing, and hopefully ATLAS as well in the future 12 OU DOE Review

13 OU Network OU connected at 10 Gbps to NLR and I2 via OneNet / GPN OU Campus back bone at 10 Gbps 10 Gbps connection straight from SRTC 3-4 Gbps from Tier 2 cluster to BNL Still some issues with asymmetric routing and throughput BNL-OU 13 OU DOE Review

14 Summary and Outlook OUHEP continues to be major contributor to ATLAS and DØ computing, as well as OSG Particularly many aspects of Grid and Distributed computing Many things accomplished Much more to do Grow Tier 2 cluster by factor of two each year; start with storage hopefully by the end of this month This will require more man power and more hardware funding 14 OU DOE Review

15 Summary and Outlook (cont.) Start utilizing OSCER resources better (for both ATLAS and DØ Computing) Aid expansion of OU Condor pool to open up more resources Continue to make make major contributions to the development, integration, and deployment of the grid middleware and ATLAS software I.e., OSG integration, deployment, monitoring, and accounting, as well as ATLAS distributed production and data management All achievable, but continued personnel and hardware funding crucial 15 OU DOE Review