LHCb Computing Model and updated requirements John Harvey

LHCb Computing model and updated requirements Slide 2 LHC Computing Review qThree technical review panels ãWorld wide analysis ãSoftware ãManagement and resources qFirst meeting of each panel took place end March qLHCb presentations available on web ã qDraft LHCb Technical Notes being prepared (see web) ãBaseline Computing model ãAnswers to SPP questions qInvite feedback qNext meetings April 14 th (Panel 3), May 12 th (Panels 1,3)

LHCb Computing model and updated requirements Slide 3 Panel 1 : World-wide analysis qGive a description of baseline computing model, indicating how it differs from MONARC generic model. qWhat is the current tentative distribution of the different regional centres? qGive a tentative assessment for data storage and management and for processing power required at the different centres (now and at the different stages towards LHC start-up). qExplain your current understanding of your policy regarding magnetic tapes versus disks. qGiven the current status of the implementation of the above model, is the experiment involved/interested in the current GRID test activities? qDo you plan Mock Data Challenges or equivalent test-beds to validate your model and provide intermediate services between now and LHC start-up? qWhat are the resources available/planned for the implementation of the above model (again staggered between now and LHC start-up)?

LHCb Computing model and updated requirements Slide 4 Panel 3 : Management and Resources qCompare computing needs in terms of hardware, software, organization, and manpower, to produce a plan to be presented to the funding agencies. qProduce an "Interim version of the Memorandum of Understanding “ indicating responsibilities of institutions ãInclude responsibilities for writing and maintenance of the software as has been done for construction of the detector. qFirst meeting focused on first 6 months of data-taking: ãPrepare a description of the computing systems needed to calibrate the subdetectors and the related strategy to get from raw data to first "simple" physics analysis. q"world wide physics analysis" left to a following meeting.

LHCb Computing model and updated requirements Slide 5 RAW Data DAQ system L2/L3 Trigger Calibration Data Reconstruction Event Summary Data (ESD) Reconstruction Tags Detector RAW Tags L3YES, sample L2/L3NO ESD Reconstruction TagsAnalysis Object Data (AOD)Physics Tags First Pass Analysis Physics Analysis Private Data Analysis Workstation Physics results ESD RAW Dataflow Model

LHCb Computing model and updated requirements Slide 6 Real Data Storage Requirements

LHCb Computing model and updated requirements Slide 7 Real Data CPU Requirements q1 SI95 = 40 MIPS qToday 1 PC ~ 10 SI95 qBy SI95/cpu (low cost), 250 SI95 (high end servers)

LHCb Computing model and updated requirements Slide 8 User Analysis Requirements qAssume that physicist performs a production analysis and requires a response time of 4 hours qThe ~10 7 events tagged by first pass analysis are scanned and candidates selected (0.25 SI95 /event) qThe selected candidates are subjected to analysis algorithm (20 SI95 / event) qTotal installed cpu power needed calculated assuming: ã~140 physicists actively doing analysis ãeach submits 1 job / day ( NB. many short jobs as well) ãanalysis distributed over a number of regional centres (~5) and assume ~20 physicists at each Regional Centre, ~40 at CERN ãAssume 0.3 x 10 7 events selected for algorithm on average ã10,000 SI95 at each Regional Centre, 20,000 SI95 at CERN

LHCb Computing model and updated requirements Slide 9 Simulation Requirements - Signal Events  CPU power to simulate 10 7 B -> D*  events in 1 year ãassume need to simulate 10 times real data sample (10 6 ) ãinstalled capacity needed is 100,000 SI95

LHCb Computing model and updated requirements Slide 10 Simulation Requirements - Background q10 5 bb inclusive events in detector every second q~100 Hz are recorded in real data ãtrigger efficiency qIf need as many to be simulated then need to generate, track, digitise and trigger bb inclusive events/yr and 10 9 will have to be reconstructed ãcorresponds to SI95 sec/yr i.e.10,000,000 SI95 qObviously need to study ways of optimising background simulation ãstore and reuse data produced at generator level (storage!) ãoptimise generation step without biasing physics ãfocus on background particularly dangerous for a specific physics channel ãaim to reduce requirements by > 1 order of magnitude qAssume 400,000 SI95 required

LHCb Computing model and updated requirements Slide 11 Simulation Requirements - Summary qFor comparison the experiments quote ãALICE (2,000,000), ATLAS/CMS (1,500,000)

CPU for production Mass Storage for RAW, ESD AOD, and TAG Institute Selected User Analyses Institute Selected User Analyses Regional Centre User analysis Production Centre Generate raw data Reconstruction Production analysis User analysis Regional Centre User analysis Regional Centre User analysis Institute Selected User Analyses Regional Centre User analysis Institute Selected User Analyses CPU for analysis Mass storage for AOD, TAG CPU and data servers AOD,TAG real : 80TB/yr sim: 120TB/yr AOD,TAG 8-12 TB/yr

Production Centre (x1) Regional Centre (~x5) Institute (~x50) Real DataSimulated Data Data collection Triggering Reconstruction Final State Reconstruction CERN Data shipped to each RC: AOD and TAG datasets 20TB x 4 times/yr= 80TB/yr User Analysis Data shipped to each Institute: AOD and TAG for samples 1TB x 10 times/yr= 10TB/yr RAL, Lyon,... Event Generation GEANT tracking Reconstruction Final State Reconstruction Data shipped to each RC: AOD, Generator and TAG datasets 30TB x 4 times/yr= 120TB/yr User Analysis Selected User Analysis Data shipped to each institute: AOD and TAG for samples 3TB x 10 times/yr= 30TB/yr

LHCb Computing model and updated requirements Slide 14 Computing at CERN qRun high level triggers and reconstruction on same cpu farm located at LHCb pit qSend RAW and ESD data over the CDR link (80 Gbps) to computer centre for archive, and first pass analysis qMaintain local storage at pit in case CDR down ãaccumulate 2 TB/day, therefore need >10 TB qDispatch AOD and TAG etc to regional centres qDuring shutdowns, down periods do re-processing of RAW data on the farm in the pit ãread RAW back from computer centre ãsend new ESD from pit to computer centre ãfull farm available so proceeds at twice the rate ãallows 2-3 reprocessings of complete year’s data qFlexible, efficient and maintainable solution

LHCb Computing model and updated requirements Slide 15 Compute Facilities at CERN DAQ / Calibration L2/L3 trigger processing Reconstruction Re-reconstruction (2-3 times per year) CDR 80 Gb/s Experiment - LHC Pit 8 CERN Computer Centre Temporary store (10 TB) for Raw Data and ESD Calibration (5TB) Data Storage and Analysis Physics Analysis MSS Readout Network Data Taking 200 Hz Raw data 20 MB/s ESD data 20 MB/s SHUTDOWN 600 Hz Raw data 60 MB/s ESD Data 60 MB/s disk RAW ESD AOD TAG Data and CPU server CPU farm AOD TAG to Regional Centres

LHCb Computing model and updated requirements Slide 16 Facility at Pit - Requirements

LHCb Computing model and updated requirements Slide 17 CERN Computer Centre Requirements

LHCb Computing model and updated requirements Slide 18 Simulation requirements q ã10 7 simulated events/yr for detector optimisation studies ãprepare TDRs q ã events/yr for high level trigger studies q ãstart to install and commission large scale facilities ãstart to produce large samples of background events with the final detector description ã~10 8 simulated events/yr q>2001 ãuse simulation and MDC to test computing model ãcontribute to HEP Application WP of EU grid proposal

LHCb Computing model and updated requirements Slide 19 Sizing Estimate for Regional Centre

LHCb Computing model and updated requirements Slide 20 Assignment of responsibility qUnderstood in LHCb institutes building subdetectors also take responsibility for development and maintenance of software qThe detector TDRs are in preparation now qMOU after TDRs

LHCb Computing model and updated requirements Slide 21 Discussion qHow will our computing needs evolve between now and 2005? qWhat regional centres will LHCb use for satisfying these needs? (RAL, CCIN2P3/Lyon, ++…) qWhat resources (cpu, storage) will be available for satisfying our simulation needs? qWhat is our attitude towards making an MOU for computing? Including software? What timescale? qWhat level of engagement should we take in Grid Projects?

LHCb Computing model and updated requirements Slide 22 GRID LHCb WP Physics Study  The total sample of B > J  /K s simulated events needed is ~10 times the number produced in the real data. qIn one year of datataking we expect to collect and fully reconstruct 10 5 events, therefore need 10 6 simulated events. qThe number of events that have to be generated, stored and reconstructed to produce this sample is q10% of the ESD data copied for systematic studies (~100 GB). qThe total amount of data generated in this production would be : RAW data 200 kB/event x 10 7 = 2.0 TB Generator data 12 kB/event x 10 7 = 0.12 TB ESD data 100 kB/event x 10 7 = 1.0 TB AOD data 20 kB/event x 10 7 = 0. 2 TB TAG data 1 kB/event x 10 7 = 0.01 TB

LHCb Computing model and updated requirements Slide 23 Grid LHCb WP - Grid Testbed qMAP farm at Liverpool has 300 processors, would take 4 months to generate the full sample of events qAll data generated (~3TB) would be transferred to RAL for archive (UK regional facility). qAll AOD and TAG datasets dispatched from RAL to other regional centres, such as Lyon and CERN. qPhysicists run jobs at the regional centre or ship AOD and TAG data to local institute and run jobs there. Also copy ESD for a fraction (~10%) of events for systematic studies (~100 GB). qThe resulting data volumes to be shipped between facilities over 4 months would be as follows : Liverpool to RAL 3 TB (RAW ESD AOD and TAG) RAL to LYON/CERN/… 0.3 TB (AOD and TAG) LYON to LHCb institute 0.3 TB (AOD and TAG) RAL to LHCb institute 100 GB (ESD for systematic studies)