Presentation is loading. Please wait.

Presentation is loading. Please wait.

Gridifying the LHCb Monte Carlo production system

Similar presentations


Presentation on theme: "Gridifying the LHCb Monte Carlo production system"— Presentation transcript:

1 Gridifying the LHCb Monte Carlo production system
Eric van Herwijnen, CERN Tuesday, 19 february 2002 Talk given at GGF4, Toronto

2 Contents LHCb LHCb distributed computing environment
Current GRID involvement Functionality of current Monte Carlo system Integration of DataGrid middleware Monitoring and control Requirements of DataGrid middleware

3 LHCb LHC collider experiment 109 events * 1Mb = 1 Pb
Problems of data storage, access and computation Monte Carlo simulation very important for detector design Need a distributed model Create, distribute and keep track of data automatically

4 LHCb distributed computing environment
15 countries, 13 European + Brazil, China, 50 institutes Tier-0: CERN Tier-1: RAL, IN2P3 (Lyon), INFN (Bologna), Nikhef, CERN + ? Tier-2: Liverpool, Edinburgh/Glasgow, Switzerland + ? (grow to ~10) Tier-3: 50 throughout collaboration Ongoing negotiatons for centres Tier-1/2/3: Germany, Russia, Poland, Spain, Brazil

5 Current GRID involvement
EU DataGrid project (involves HEP, Biology, Medecine and Earth Observation sciences) Active in WP8 (HEP applications) of DataGrid Use “middleware” (WP1-5) + Testbed (WP6) + Network (WP7) Current distributed system works since some time, LHCb is: Grid enabled, but not Grid dependent

6 MC production facilities (summer 2001)
Centre Max. (av.) # of CPUs available simultaneously Batch System Typical weekly production % submitted through GRID CERN 315 (60) LSF 85 k 10% RAL 100 (60) PBS 35k 100% IN2P3 225 (60) BQS Liverpool 300 (250) Custom 150k 0% Bologna 20 (20) Nikhef 40 (40) Bristol 10 (10) 15k

7 Submit jobs remotely viaWeb Update bookkeeping database Transfer data to Mass store Execute on farm Monitor performance of farm via Web Data Quality Check

8 GRID-enabling production
Run mc executable write log to Web copy data to mass store (dg-data-copy) call CERN servlet Construct job script and submit via Web (dg- authentication, dg-job-submit) mass store FTP servlet (dg-data-replication) copy data to CERN mass store call servlet to copy data from local mass store to CERN update bookkeeping db (?LDAP-now Oracle)

9 Gridi-fying the MC production system
Provide a convenient tool for DataGrid Testbed validation tests Feed back improvements into the MC system currently in production Clone current system, replace commands by DataGrid middleware Report back to WP8 and other workpackages as required

10 Monitoring and control of running jobs
Control system to monitoring distributed production (based on PVSS, author: Clara Gaspar) Initially for MC production, later all Grid computing Automatic quality checks on final data samples Online histograms and comparisons between histograms Use DataGrid monitoring tools Feed back improvements into production MC system

11

12 Requirements on DataGrid middleware
Security: single user logon Job submission: use “sandboxes” to package environment so that use of AFS is unnecessary Monitoring: integrate with WP3 tools where possible for farm monitoring, use own tools for data quality monitoring Data moving: use a single API to move data We are in a cycle of requirements, design, implementation and testing


Download ppt "Gridifying the LHCb Monte Carlo production system"

Similar presentations


Ads by Google