Distributed physics analysis for ATLAS: presentation to the LHCC Comprehensive Review James R. Catmore Research Associate, Lancaster University, UK ATLAS B-physics working group
ATLAS Analysis computing model 2 TIER 0 (CERN) and PRODUCTION SYSTEM (TIER 2 SITES) Detector Digits ESD Digits Monte Carlo Reconstruction AOD & TAG building ESD AOD Jobs TAG AOD Results TAG ANALYSTS TIER 1 SITES TIER 2 SITES MIDDLEWARE James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
ATLAS distributed analysis tools 3 • Distributed data management: DQ2 ‣ Replica catalogue providing mappings to concrete datasets on sites across the Grids ‣ Dataset: basic data unit for an analyst - may contain thousands of files ‣ Tools (command line and web) for listing catalogue content • Metadata service: AMI ‣ Web interface giving information on file provenance • Grid user interface: GANGA James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
GANGA 4 • Grid user interface: GANGA ‣ Single tool for all Grid-based work (including analysis and small Monte Carlo productions) ‣ Trivial switching between Grid running and local execution (for testing purposes) ‣ Grid backends include LCG (EGEE), NorduGrid, OSG (Panda) ‣ Interface via a command-line or a GUI www.cern.ch/ganga James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
TAGs 5 • Event TAGs are the method by which reconstructed events are selected for analysis • Built from AOD according to offline analysis-style code ‣ Initially consists of files containing event metadata and a pointer to the POOL file from which the tag was made ‣ Later loaded into a relational database for access by physicists • Typical information held in the TAG: temporal conditions, quality & detector status, trigger, physics James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Analysts’ work-flow (data taking) 6 1 Prepare analysis code Prepare TAG selection 2 Set up analysis job 3 Submit to the Grid 4 Retrieve results 5 Merge results 6 Inspect in ROOT 7 James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Analysts’ work-flow (now) 7 1 Prepare analysis code Locate dataset 2 Set up analysis job 3 Submit to the Grid 4 Retrieve results 5 Merge results 6 Inspect in ROOT 7 James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Example for this presentation 8 • Submission of a simple Athena analysis job to the Grid using GANGA • Search for J/ ψ→μμ decays in AOD using the Athena package BPhysAnalysisTools • Run over Monte Carlo AOD data sample produced by the Production System • Job goes to the data • Physics analyses using exactly this method are in progress now for commissioning preparation James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Step 1: prepare your analysis code 9 James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
☝ Step 2: Find your dataset in AMI (current) 10 Dataset Wildcards Simulation chain step number supported All official ATLAS MC data on the Grid uses a strict naming policy AMI page: http://ami3.in2p3.fr:8080/opencms/opencms/AMI/www James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Step 2: Find your dataset in AMI (current) 11 Dataset Data Athena Status DQ2 link name type release James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Step 2: prepare TAG selection (data taking) 12 • Event Quantities • Data Quality • Trigger information • Electron objects • Photon objects Available for TAG selection • Muon objects • Tau-jet objects • Jets • Physics attributes Developing the correct selection will be a major task and will probably involve several analysts or an entire physics group. Local testing will be required (local use of the TAGs is trivial) James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Step 3: Set up analysis job.... GANGA window 13 James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Step 3: Set up analysis job.... GANGA window 14 Retrieve New Save Kill output Job control panel Open Submit Copy Delete Job details window James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Step 3: Set up analysis job.... GANGA window 15 Job monitoring window Type Back-end ID Status James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Step 3: Set up analysis job.... job builder 16 Athena Back-end control Datset/TAG Results James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Where is the data.....? 17 Best sites are FZKDISK and LYONDISK: 592 files each James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Where is the data? 18 James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
☝ Step 4: Submit your job.... 19 James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Step 6: Monitor the job 20 after 5 minutes or so..... James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
☝ Step 7: Retrieve and merge your results 21 Retrieve Merge James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Step 8: Collect results; analysis in ROOT 22 James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Producing Monte Carlo data in GANGA 23 • Physicists will often need to produce quick small samples of Monte Carlo data (<10 000 events) • GANGA provides a plug-in for running the full Athena simulation chain • Uses the standard ATLAS JobTransformation mechanism used in the main production system to guarantee trustworthiness • Naming abides by ATLAS conventions • User-generated datasets saved on the Grid and registered in DQ2 under the user’s name James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Monitoring 24 ATLAS dashboard - reached from GANGA web page James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Conclusions and outlook 25 • ATLAS physicists are doing distributed physics analysis now on Monte Carlo data • TAGs are under preparation for data-taking • Datasets are well documented in AMI but even replication across the sites remains a problem ‣ The submission tools are sufficiently robust to handle this • Heavy reliance on LXPLUS UI at the moment • How will user support work when data-taking begins? • ATLAS distributed computing will be ready for data James Catmore LHCC Comprehensive Review, CERN, 20th November 2007
Recommend
More recommend
