1. 1 LHCb Computing A.Tsaregorodtsev, CPPM, Marseille 14 March 2007, Clermont-Ferrand

2. 2 LHCb in brief  Experiment dedicated to studying CP-violation  Responsible for the dominance of matter on antimatter  Matter-antimatter difference studied using the b-quark (beauty)  High precision physics (tiny difference…)  Single arm spectrometer  Looks like a fixed-target experiment  Smallest of the 4 big LHC experiments  ~500 physicists  Nevertheless, computing is also a challenge….

3. 3 LHCb Basic Computing principles  Raw data shipped in real time to Tier-0  Registered in the Grid (File Catalog)  Raw data provenance in a Bookkeeping database (query- enabled)  Resilience enforced by a second copy at Tier-1’s  Rate: ~2000 evts/s (35 kB)  70 MB/s  4 main trigger sources (with little overlap) b-exclusive; dimuon; D*; b-inclusive  All data processing up to final Tuple or histogram production distributed  Not even possible to reconstruct all data at Tier0…  Processing at Tier1 centers  LHCb runs jobs where data are  All data are placed explicitly

4. 4 LHCb data processing software Simul. Gauss Analysis DaVinci MCHits DST Raw Data (r)DST MCParts GenParts Event model / Physics event model AOD Conditions Database Gaudi Digit. Boole Trigger Moore Recons. Brunel

5. 5 LHCb dataflow Online MSS-SE Tier1 MSS-SE Recons. Stripping Simulation. Raw Digi Raw/Digi rDST DST rDST+Raw Tier1 Tier2Tier0 Analysis DST

6. 6 Computing Model

7. 7 The LHCb Tier1s  6 Tier1s  CNAF (IT, Bologna)  GridKa (DE, Karlsruhe)  IN2P3 (FR, Lyon)  NIKHEF (NL, Amsterdam)  PIC (ES, Barcelona)  RAL (UK, Didcot)  Contribute  Reconstruction  Stripping  Analysis  Keeps copies on MSS of  Raw (2 copies shared)  Locally produced rDST  DST (2 copies)  MC data (2 copies)  Keeps copies on disk of  DST (7 copies)

8. 8 LHCb Computing: a few numbers  Event sizes  on persistent medium (not in memory)  Processing time  Best estimates as of today  Requirements for 2008  4 10 6 seconds of beam Current estimate TDR estimate 0.3 0.2 2.4 100 25 0.3Analysis 0.2Stripping 2.4Reconstruction kSI2k.sEvt processing 110DST 20rDST 35RAW kBEvent Size

9. 9 Summary resources needs for 2008 775 Stripped 1143754376 AnalysisSimulationrDSTRAW Data on disk 2008 (TB) 0.5 Stripping 0.54.61.4 AnalysisSimulationRecons. CPU needs in 2008 (MSI2k.yr) 483 Stripped -128320560 AnalysisSimulationrDSTRAW Data on tape 2008 (TB)

10. 10 DIRAC grid management software  DIRAC is a distributed data production and analysis system for the LHCb experiment  Includes workload and data management components Uses LCG services whenever possible  Was developed originally for the MC data production tasks  The goal was: integrate all the heterogeneous computing resources available to LHCb Minimize human intervention at LHCb sites  The resulting design led to an architecture based on a set of services and a network of light distributed agents

11. 11 DIRAC Services, Agents and Resources DIRAC Job Management Service DIRAC Job Management Service Agent Production Manager Production Manager GANGA DIRAC API JobMonitorSvc JobAccountingSvc Job monitor ConfigurationSvc FileCatalogSvc BookkeepingSvc BK query webpage BK query webpage FileCatalog browser FileCatalog browser Services Agent MessageSvc Resources LCG Grid WN Site Gatekeeper Tier1 VO-box

12. 12 WMS Service  DIRAC Workload Management System is itself composed of a set of central services, pilot agents and job wrappers  Realizes the PULL scheduling paradigm  Pilot agents deployed at LCG Worker Nodes pull the jobs from the central Task Queue  The central Task Queue allows to apply easily the VO policies by prioritization of the user jobs  Using the accounting information and user identities, groups and roles  The job scheduling is late  Job goes to a resource for immediate execution

13. 13 Task Queue 1 DIRAC workload management Job Receiver Job Database Optimizer Prioritizer Optimizer Data Optimizer XXX Priority Calculator Accounting Service LHCb policy, quotas Job requirements, ownership Job priority Task Queue 1 VOMS info Agent Director Agent 1 Agent 2 … Match Maker Resources (WNs) Central Services

14. 14 VO-boxes  LHCb VO-boxes are machines offered by Tier1 sites to insure safety and efficiency of the grid operations  Standard LCG software is maintained by the site managers;  LHCb software is maintained by the LHCb administrators;  Recovery of failed data transfers and bookkeeping operations.  VO-boxes are behaving in a completely non-intrusive way  Access site grid services via standard interfaces  Main advantage – geographical distribution  VO-boxes are set up now in all the Tier1 centers.  Any job can set requests on any VO-box in a round-robin way for redundancy and load-balancing

15. 15 DM Components  DIRAC Data Management tools are built on top of or provide interfaces to the existing services  The main components are:  Storage Element client and Storage access plug-ins SRM, GridFTP, HTTP, SFTP, FTP, …  Replica Manager – high level operations Uploading, replication, registration Best replica finding Failure retries with alternative data access methods  File Catalogs LFC Processing Database  High level tools for automatic bulk data transfers  T0-T1 raw data distribution  T1-T1 reconstructed data distribution

16. 16 Bulk Data Management  Bulk asynchronous file replication  Requests set in RequestDB  Transfer Agent executes periodically  ‘Waiting’ or ‘Running’ requests obtained from RequestDB  FTS bulk transfer jobs submitted and monitored LCG – Machinery Transfer network LHCb - DIRAC DMS Request DB FC Interface Transfer Manager Interface Replica Manager Transfer Agent LCG File Catalog File Transfer Service Tier0 SE Tier1 SE A Tier1 SE B Tier1 SE C

17. 17 Results: T0-T1 data transfer tests

18. 18 Processing Database  The suite of Production Manager tools to facilitate the routine production tasks:  define complex production workflows  manage large numbers of production jobs  Transformation Agents prepare data reprocessing jobs automatically as soon as the input files are registered in the Processing Database via a standard File Catalog interface  Minimize the human intervention, speed up standard production

19. 19 DIRAC production performance  Permanent production (no more DC)  Up to 10K simultaneous production jobs  The throughput is only limited by the capacity available on LCG  ~80 distinct sites accessed through LCG or through DIRAC directly IN2P3 Total

20. 20 CPU usage by country (since Dec. 2006) CountryCPU usage (%) UK41.1 CERN12.1 Italy9.6 German8.1 France7.7 Spain6.6 Greece3.8 Netherland3.1 Polony2.4 Russia2.0 Hungary0.9 UK CERN German Spain France Italy

21. 21 CPU usage since April 2006 by sites Sites IDCPU usage (%) CERN14.0 CNAF11.1 Manchester9.1 RAL7.4 GRIDKA4.9 QMUL4.7 IN2P34.5 LPC3.3 USC3.3 NIKHEF2.9 Brunel(uk)2.4 Barcelona2.0 Liverpool1.9 Glasgow1.7 HG-06(Greece)1.6 Lancashire1.4 PIC1.3 CERN CNAF Manchester RAL  50% of production done at T1s

22. 22 Status of various activities  The LHCb production system is working in a stable way for the MC production  Pilot agent model screens the LCG inefficiencies  The data distribution and reprocessing is more difficult  Unstable storage systems  Flaws in the data access middleware  Many problems will be resolved (and created) with the new SRM2.2 release

23. 23 Status of various activities  User analysis is starting  Reliable data access is a crucial point  Efficient job prioritization is a must User and production jobs are competing for the same resources  Full chain tests involving DAQ, T0-T1 data distribution, distributed reconstruction, T1-T1 data distribution, final analysis in June  Automatic “real time” data movement and processing  Close to a real scale involving all the T1 sites

24. 24 Conclusions  LHCb has proposed a Computing Model adapted at its specific needs (number of events, event size, low number of physics candidates)  Reconstruction, stripping and analysis resources located at Tier1s (and possibly some Tier2s with enough storage and CPU capacities)  CPU requirements dominated by Monte-Carlo, assigned to Tier2s and opportunistic sites  With DIRAC, even idle desktops / laptops could be used ;-) LHCb@home ?  Requirements are modest compared to other experiments  DIRAC is well suited and adapted to this computing model  Integrated WMS and DMS  LHCb’s Computing should be ready when first data come

25. 25 LHCb software stack  Uses CMT for build and configuration (handling dependencies)  LHCb projects:  Applications Gauss (simulation), Boole (digitisation), Brunel (reconstruction), Moore (HLT), DaVinci (analysis)  Algorithms LBCOM (commone packages), Rec (reconstruction), Phys (physics), Online  Event model LHCb  Software framework Gaudi  LCG Applications area POOL, root, COOL  Lcg/external External SW: boost, xerces… also middleware client (lfc, gfal,…) LHCb Online SEALPOOL RootExt.Libs GaussBooleBrunelPanoramixMoore Gaudi LCG Framework Applications PhysRecLbcom Event Model DaVinci Component projects COOL CORAL Geant4 GENSER

26. 26 Last month activities Record of running jobs 9654 CNAFGRIDKA IN2P3 NIKHEF PIC RAL ALL CERN  Average of 7.5K running jobs in the last month  Temporary problems at PIC and RAL

27. 27 Community Overlay Network TQ DIRAC WMS WN Pilot Agent Pilot Agent Monitoring Logging WN Pilot Agent Pilot Agent WN Pilot Agent Pilot Agent  DIRAC Central Services and Pilot Agents form a dynamic distributed system as easy to manage as an ordinary batch system  Uniform view of the resources independent of their nature Grids, clusters, PCs  Prioritization according to VO policies and accounting  Possibility to reuse batch system tools  E.g. Maui scheduler GRID

28. 28 Reconstruction requirements 2 passes per year: 1 quasi real time over ~100 day period (2.8 MSI2k) re-processing over 2 month period of shutdown (4.3 MSI2k) Make use of Filter Farm at pit (2.2 MSI2k) - data back to the pit 1.520.680.230.450.15CPU (MSI2k.yr) 16072244816MSS storage (TB) 8  10 9 3.6  10 9 1.2  10 9 2.4  10 9 8  10 8 Number of events 1.00.450.150.30.1Input fraction Totalb-inclusiveD*Dimuonb-exclusive

29. 29 Stripping requirements Stripping 4 times per year - 1 month production outside of recons Stripping has at least 4 output streams Only rDST stored for “non-b” channels+RAW i.e. 55 kB RAW+full DST for “b” channels - i.e. 110kB Output on disk SE at all Tier-1 centres 84121TAG (TB) 52413269Storage requirement per stripping (TB) 0.110.020.030.060.02CPU (MSI2k.year) 8.4  10 9 3.6  10 7 2.4  10 8 4.8  10 8 8  10 7 Event yield per stripping 9.571005510Reduction factor 1.000.450.150.30.1Input fraction TotalInclusive-bD*dimuonExclusive-b

30. 30 Simulation requirements - studies to measure performance of detector & event selection in particular regions of phase space - use large statistics dimuon & D* samples for systematics - reduced Monte Carlo needs 3.87Total 0.012.48  10 7 Brunel 0.0318  10 8 Boole 1.9758  10 8 GaussInclusive 0.012.48  10 7 Brunel 0.0318  10 8 Boole 1.9758  10 8 GaussSignal Total CPU (MSI2k.year) CPU time/evt (kSI2k.s) Nos. of eventsApplication

31. 31 Simulation storage requirements - Simulation still dominate LHCb CPU needs - Current evt size for Monte Carlo DST (with truth info) is ~400kB/evt; - Total storage needs 64TB in 2008 - Output at CERN and another 2 copies distributed over Tier-1 centres 64Total 0.118  10 7 TAG 324008  10 7 DSTInclusive 0.118  10 7 TAG 324008  10 7 DSTSignal Total Storage (TB) Storage/evt (kB) Nos. of eventsOutput

32. 32 Analysis requirements - user analysis accounted in model predominantly batch - ~30k jobs/year - predominantly analysing ~10 6 events - CPU of 0.3 kSI2k.s/evt - Analysis needs grow linearly with year in early phase of expt 802008 Disk storage (TB) 0.312008 CPU needs (MSI2k.years) 10Number of “active” Ntuples 5Event size reduction factor after analysis 4Nos. of analysis jobs per physicist/week 140Nos. of physicist performing analysis