LHCb Computing activities Philippe Charpentier CERN – LHCb On behalf of the LHCb Computing Group

LHCb Computing activities PhC2 Summary of Computing activities m Simulation o Mainly used for identifying background and evaluating acceptances and efficiencies o Simulates an ideal detector, however with realistic geometry o Event generation and detector response tuned to real data P Iterative process, depends on the data taking yesr m Real data handling and processing o Distribution to Tier1s (RAW) o Reconstruction (SDST) o Stripping and streaming (DST) o Group-level production (µDST) m User analysis o MC and real data processing o Detector and efficiency calibration o End-user analysis (usually off-Grid: Tier3 or desktop)

LHCb Computing activities Simulation jobs m 5 steps jobs m Gauss: simulation, based on Geant4 m Boole: digitisation m Moore: trigger m Brunel: reconstruction m DaVinci: stripping (single stream) m Any file may be saved, usually only the final DST (currently uploaded to CERN) m 100 to 200 events per job m 5 to 10 hours duration m 40 to 80 MB per file m Merging required (see later) PhC3 SIM DIGI DST (X)GEN Gauss Boole Brunel DaVinci DIGI Moore

LHCb Computing activities Reconstruction-Stripping jobs m 2 steps jobs m One input file: copy to local disk m Brunel: reconstruction m DaVinci: stripping and streaming o Around 10 to 13 streams m SDST and DSTs saved (locally) o SDST: local T1D0 (LHCb_RDST) o (µ)DSTs: temporary T0D1 (LHCb_DST) m RAW files up to 3 GB (45,000 events) m 2s per event: 1 day jobs m SDST: 80% of RAW size m All DSTs: ~10% of RAW size o Individual files small (10’s MB) PhC4 SDST DST Brunel DaVinci DST µDST RAW

LHCb Computing activities Stripping jobs m One step jobs m Multiple input files (e.g. 10 SDSTs) o RAW files required as well o All files must be present on disk cache (possibly staged) o Access by protocol (xroot, dcap…) m DSTs 10 times bigger than for Reco-Stripping o Saved locally (LHCb_DST) o Streamed DSTs ~ 100’s MB o Merging still required PhC5 SDST DST DaVinci DST µDST RAW

LHCb Computing activities Merging jobs m Automatically generated from job output o Simulation o Reco-stripping or Stripping m Typically 5 GB files o For real data, only merge data from the same run o Run duration can easily be adjusted online P was 1 hour in 2010 (7 Mevts, ~200 RAW files of 3 GB) m Uses download policy for input data o Limited by number of files: P SRM overload P SE and disk overload o See Ricardo’s talk tomorrow m Merged files uploaded locally to T1D1 (master copy) o LHCb_M-DST or LHCb_MC_M-DST PhC6

LHCb Computing activities Data replication m Performed by a data driven transformation o Same mechanism as productions P See Andrew Smith’s talk at CHEP10 m Distribution policy implements the LHCb Computing Model o RAW files: one Tier1 from the Tier0 replica o MC: 2 master replicas (CERN + Tier1), 1 secondary replica P Sites selected randomly P Foresee to implement space driven policy o Real data: 2 master replicas (CERN + Tier1), 2 secondary replicas P Differs from CM (should be 5 secondary replicas) P Adaptation following larger event sizes P Each run is distributed to the same sites m Replication using FTS PhC7

LHCb Computing activities RAW streams 2010 m In 2010, two RAW streams: o EXPRESS: selected small rate (typically 5 Hz) P Used for Data Quality checking P Smaller files (200 MB), kept at CERN only P CERN, quasi online (less than 2 hours) P DQ result used to allow FULL processing o FULL: all triggers, around 2,000 Hz P Distributed to Tier1s (after successful tape migration) P Using the CPU pledge share for distribution d Share set to 0 when site is unavailable P Full runs assigned to single site ( files) P Reconstruction-Stripping launched automatically after the run is declared OK P Merging performed on the same site PhC8

LHCb Computing activities RAW streams 2011 m 2011 (options): o EXPRESS as 2010 o MINBIAS: around 200 Hz of minimum/no bias triggers P Special stream, treated like the FULL stream o b-physics triggers and charm-physics. Options: P 3,000 Hz on FULL stream P 2,000 Hz on BPHYS stream, 1,000 Hz on CHARMPHYS stream P To be decided soon P Replication to Tier1s as for FULL P Staged processing for CHARMPHYS triggers (depending on CPU availability) with lower priority m Processing o Run Reco and Stripping in separate jobs o Alleviates the problem of large number of files to be merged P Use SDSTs as input to stripping jobs P Caveat: use protocol access, SRM possibly overloaded at start of jobs (get tURLs at once) PhC9

LHCb Computing activities Additional possible changes m Reconstruction at Tier2: o Using files download from a well connected Tier1 o Selected number of Tier2s o Better control than Analysis P All handled by the Production team, better organised, less chaotic o Most probably needed for end-of-year reprocessing m Analysis at Tier2s: o Still to be commissioned P Do we need it? o Problem of local manpower and storage management m LHCb prefers reconstruction to analysis o For the time being, no limitation seen for user jobs at Tier1s P Turn-around still very good (all user jobs finished overnight) PhC10