ATLAS HLT/DAQ V. Vercesi for the ATLAS Italia HLT/DAQ Group CSN1 Trieste Settembre 2006
V. Vercesi - INFN Pavia 2 S. Falciano (Roma1) Coordinatore Commissioning HLT A. Negri (Irvine, Pavia) Coordinatore Event Filter Dataflow A. Nisati (Roma1) TDAQ Institute Board chair e Coordinatore PESA Muon Slice F. Parodi (Genova) Coordinatore b-tagging PESA V. Vercesi (Pavia) Deputy HLT leader e Coordinatore PESA (Physics and Event Selection Architecture) Attività italiane Trigger di Livello-1 muoni barrel (Napoli, Roma1, Roma2) Trigger di Livello-2 muoni (Pisa, Roma1) Trigger di Livello-2 pixel (Genova) Event Filter Dataflow (LNF, Pavia) Selection software steering (Genova) Event Filter Muoni (Lecce, Napoli, Pavia, Roma1) DAQ (LNF, Pavia, Roma1) DCS (Napoli, Roma1, Roma2) Monitoring (Cosenza, Napoli, Pavia, Pisa) HLT/DAQ system commissioning and exploitation (Everybody)
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 3 ATLAS TDAQ Dual(quad)-CPU nodes SDX1 USA15 UX15 ATLAS detector Read- Out Drivers ( RODs ) First- level trigger Read-Out Subsystems ( ROSs ) UX15 USA15 Dedicated links Timing Trigger Control (TTC) 1600 Read- Out Links Gigabit Ethernet RoI Builder Regions Of Interest VME ~150 PCs Data of events accepted by first-level trigger Event data requests Delete commands Requested event data Event data ≤ 100 kHz, 1600 fragments of ~ 1 kByte each LVL2 Super- visor DataFlow Manager Event Filter (EF) pROS ~ 500~1600 stores LVL2 output ~100~30 Network switches Event data pulled: partial ≤ 100 kHz, full ~ 3 kHz Event rate ~ 200 Hz Data storage Local Storage SubFarm Outputs (SFOs) LVL2 farm Network switches Event Builder SubFarm Inputs (SFIs) Second- level trigger
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 4 Efficienza dell’algoritmo di selezione è > 99% Tutte le inefficienze sono dovute a zone dello spettrometro non coperte da RPC. In particolare: Settori speciali (Feet), ascensore, crack =0 e supporti del magnete. Efficienza dell’algoritmo di selezione è > 99% Tutte le inefficienze sono dovute a zone dello spettrometro non coperte da RPC. In particolare: Settori speciali (Feet), ascensore, crack =0 e supporti del magnete. No RPC hit No Trigger Low-Pt Trigger DC3 data DC2 data Muon sources 6 GeV 6 GeV threshold Lumi= GeV 20 GeV threshold Lumi=10 34 /K b c W327 tNegligible Low-pT 6 GeV Threshold Efficienza globale Level-1 nel Barrel: 83% Low-pT Trigger 79% High-pT Trigger Level-1 Muon Trigger System
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 5 One Switch rack - TDAQ rack port GEth for L2+EB One ROS rack - TC rack + horiz. Cooling - 12 ROS 48 ROBINs One Full L2 rack - TDAQ rack - 30 HLT PCs Partial Superv’r rack - TDAQ rack - 3 HE PCs Partial EFIO rack - TDAQ rack - 10 HE PC (6 SFI - 2 SFO - 2 DFM) Partial EF rack - TDAQ rack - 12 HLT PCs Partial ONLINE rack - TDAQ rack - 4 HLT PC (monitoring) 2 LE PC (control) 2 Central FileServers RoIB rack - TC rack + horiz. cooling - 50% of RoIB 5.5 surface: SDX1 underground : USA15 Pre-series system in ATLAS point-1 8 racks (10% of final dataflow, 2% of EF) ROS, L2, EFIO and EF racks: one Local File Server, one or more Local Switches Machine Park: Dual Opteron and Xeon nodes, uniprocessor ROS nodes Operating System: Net booted and diskless nodes, running SLC3
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 6 Event Building Parameterizing observed EB measurements, we better understand the conditions for stable & performant operations TS = number of requests from SFI nodes for event fragments WT = number of events processed in parallel by L2 farm = L2 accept ratio
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 7 Pre-series studies
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 8 ROD Crate DAQ Last major upgrades To accommodate detectors’ requests New API for operational monitoring To get “coherent” operation information Simultaneous module readout in a crate RCD creates a packet of info and sends it to Monitoring A monitoring task analyses the packet Further improvements in last TDAQ version OKS schema for RCD and ROS simplified Detector modules in a crate can be managed Serially or in multi-threaded mode Each state transition can be set as multi-threaded Can dramatically improve initialization time RCD plans The RCD structure should be stable enough All the main requests from detectors have been fulfilled We continue supporting the implementations Main task is now RCD commissioning Goes on with ROD commissioning Includes support to all detectors
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 9 Monitoring Framework
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 10Monitoring Reliable and stable GNAM CORE Configurable from on-line database, with histogram history Logging system (to migrate to ERS) Access Manager to ATLAS conditions DB Detector libraries MDT, RPC, Tile, CSC, (TGC): in use for commissioning Interface to Event Display available (for some detectors) Main issue: Smart Monitoring Automatically catch most low-level problems Needed for reaching and maintaining stable running conditions Common tools needed Histogram comparison toolkit Error message routing and filtering Detector experts Identify most common problem sources Implement and test alarm strategies
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 11 HLT Selection algorithms Lots of ongoing activities on algorithm development side LVL1 and LVL2 algorithms run in CSC production for all trigger slices EF algorithms almost finished (target , next week) Lots of ongoing work to improve/test/validate algorithms PESA algorithms being reviewed (Project Milestone) ID LVL2, ID EF, Calo LVL2 e/g reviews have already taken place Muon LVL2+EF ~90% complete, Calo EF e/ in October Calo jet/tau/etmiss simple extension of e/ Building up comprehensive information about performance Ongoing work for testing in athenaMT/athenaPT Measurements of system (timing) and physics performance LVL1/HLT AODs fully available in Rel 12 for trigger-aware analyses (Project Milestone) Most data in place for ESD & AOD in release , Apr 06 More hypothesis algorithms, slice configs and TriggerDecision in
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 12 HLT Steering Il cuore delle sequenze di selezione e decisione Sviluppi basati sull’attuale implementazione (12.0.2) Introduzione di selezioni topologiche elementari Primi test attualmente in sviluppo nel settore della B-fisica (B→ , J/Psi→ , B S →D S con D S → ) Introduzione di meccanismi per la sincronizzazione delle sequenze di algoritmi Selezioni combinate di diversi oggetti (ad es. singolo e, singolo jet, singolo ) Sincronizzazione delle selezioni topologiche Iterazione del design di implementazione Ridisegno dell’implementazione dello Steering Nuove interfacce, più generali e flessibili, per gli algoritmi di feature extraction e hypothesis testing Ristrutturazione della navigazione, disaccoppiata dallo Steering, per essere facilmente utilizzata nelle analisi Offline Interfacciamento con l’introduzione del concetto di Trigger Chain Fase di disegno quasi ultimata: inizio dell’attività di sviluppo e test
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 13 Electrons Eff. w.r.t truth for E T =25GeV e± Time per track in top events (~1.3 tracks/RoI with pTmin=0.5GeV) Brem recovery to be optimized: in progress… Optimizations ongoing also for eff/rej Photons Use only calorimeter information at L2 and EF Use CSC data, single photons E T = 20, 60 GeV, filtered QCD di-jets with E T (hard) > 17 GeV 97% 98% AlgorithmKalmanFitterGSF Time/track9.5 ms1.24 s Eff % 20i Rate 2 20i Eff % 60 Rate 60 L Hz Hz L2 Calo92.76 Hz Hz EF Calo85.34 Hz Hz
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 14 Efficiencies for SM signals trigger 2e15i67.2%--- e25i92.9%79.6% e6020.4%6.9% all94.8%80.3% Z ee W e Efficiencies given after kinematical cuts: 2e in | | 15 GeV for Z ee 1e in | | 25 GeV for W e Use trigger optimisations giving 80% overall eff (outside crack) for 2e15i, e25i, e60. e25i efficiency stays flat after turn-on. efficiency stays flat after turn-on. Turn on curves similar for e60. # after L1+L2+EF # after kinematical cuts eff Method for single-electron trigger efficiency determination from data in progress (, ) Z ee Z
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 15 Trigger efficiency from Z → μ + μ - Double Object (DO) method Control sample: “Good” Z from 2 offline + - with loose selection cuts + 1 ± trigger signature satisfied Trigger efficiency determined from counting in how many cases the second ± satisfies the trigger requirements EF TrigMoore (MuonSpectrometer standalone) EF TrigMoore (MuonSpectrometer + InnerDetector) RecoMC ε EF (%)96.4 ± 0.2 ε L1+EF (%)79.9 ± ± 0.2 RecoMC ε EF (%)94.5 ± 0.2 ε L1+EF (%)78.3 ± ± 0.2 For (µ20)>70% statistical uncertainty after 30min at L=10 33 cm -2 s -1: ~1-2% Ongoing study using complementary method of orthogonal signature from ID (DOS method), almost finished
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 16 Barrel: Final update of the LUT available: optimized with respect to the momentum resolution and to the efficiency Hypothesis exists and the P T cut is optimized for the 6 GeV threshold Endcap: Bugs found in the TGC reconstruction code and in the analysis programs Situation improved: harsh zones still remains, but results are understood now Only TGC data has been used: try to measure muon P T and to parameterize the track path An offline study of the momentum resolution obtained fom MDT data is also under way: preliminary results are in agreement with those we got using only TGC data. Large dishomogenity Endcap stations outside magnetic field Innermost TGC wheels have reduced coverage in small sectors EndCap 6 GeV 20 GeV Endcap: 1.05 <l l <2.4 CSC data prelim. p T (GeV) efficiency
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 17 Very active trigger slices, just one year ago only small part of the selections were available (new groups joining TDAQ) Jets Algorithms for LVL1/LVL2/EF reco and hypothesis in place Lots of work carried out to allow trigger-aware analysis Ongoing work to understand physics performance Tau Algorithms for LVL1/LVL2/EF reco and hypothesis in place Use tauRec at EF, support for other tau reco packages foreseen Ongoing work to find best sequence at LVL2: first calo reconstruction or first tracking reconstruction L2 selection ran successfully in athenaMT in ETmiss Study performance of ETmiss at LVL1 and EF
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 18 Prescale Jets l First place where to study and implement pre-scales l Reconstructed Jet E T distribution for a combination of J1 to J8 samples, scaled according to their respective cross-section
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 19 Algorithms in Pre-Serie Motivation Get the trigger PESA algorithms working online Make sure the latest offline developments get running online Historical background Tools allowing the emulation of online running were developed athenaMT (LVL2) and athenaPT (EF) A group of people was formed in order to provide full slice job options suitable for online running Jets (Ignacio Aracena) Taus (Pilar Casado, Richard Soluk) Egamma (Xin Wu, Imma Riu) Muons (Alessandro Di Mattia, Diana Scannicchio) One single integrated job options built successfully able to run in the Pre-Serie L2 and EF farms from real data format (bytestream) Try this integrated job options in an online partition with different input files Single electrons, Top events, Muon events Produce a unified input BS file to be tested online Produce a BS output file from SFO and try extending the menu to other signatures (e.g. di-objects) using the same algorithms slices to see the trigger behaviour
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 20Running L2 partition running ONLINE single electrons with the integrated job options in the nightlies of
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 21 More running
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 22Databases Online OracleDB Offline master CondDB Tier-0 recon replica Diagnostics replica Tier-1 replica Tier-1 replica Online Detectors HLT farm Diagnostics Tier-0 farm Bypass (ATCN / CERN-IT) ATLAS Point1 CERN Computer centre Remote sites Isolation / cut Calibration Updates including special Muons calibration path ATCN Network OKS in-memory OODB COOL RelDB CORAL* DCS Athena Server Data: in the DB, plus in (large) files referenced from the DB RDB systems used: Oracle, MySQL, sqlite *layer to decouple clients
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 23 Trigger Configuration TriggerTool GUI for DB population easy and consistent menu changes for experts (LVL1 and HLT) TriggerDB stores all information to configure the trigger: LVL1 menu, HLT menu, HLT algorithm parameters (JO), HLT release information stores all versions used, with a key Configuration and Conditions DB DB available at Point 1 and replicated to external sites Retrieval of information for running get information by key via two paths extraction of data in XML/JO files direct read-out for both online + offline running TriggerDB online running offline running shift crewoffline userexpert TriggerTool DB population scripts Data Flow: R/O interface Configuration System compilers
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 24 Example database for TrigConf LVL1 HLT jobOptions HLT menu HLT release keys: stored in Conditions DB => Trigger conditions precisely known
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 25 HLT Large Scale Tests Test complete HLT on many nodes Scalability of DAQ, DB, HLT sw Emphasis on DB, HLT (no LVL1) DAQ: verify scalability & stability DataBase: Explore caching tools (DBProxy) Detector oriented tests 1000 to 1200 nodes LST 2005: 600 (512 used for LVL2) Will be provided by IT (agreed) with lxbatch etc. machines Non-optimized network Big enough for full size Level2, ¼ size Event Filter Test Level-2 and Event Filter together (+EB and SFO) As many trigger algorithms as possible Large number of events, recycled Collect monitoring info Physics plots (produced by algorithms) Detailed timing histograms, etc. Need to have SHIFTS!
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 26 LVL2 ID in cosmics Tracking algorithms IdScan, modified to handle single tracks with large impact parameters SiTrack, version with special LUT tuned for muons not coming from IP Variables studied For T2Id, Offline and MC: Ntracks, d0, phi0, z0, eta0 Efficiency calculations Data: LVL2 wrt Offline Simulation: LVL2 and Offline wrt truth Cosmics at SR1: instrumented region, two SCT sectors
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 27 Trigger-aware analysis Analyses using trigger information as a “pre- processor” to correctly evaluate efficiencies, physics reach, etc. The reconstructed objects, used by the trigger are saved in the ESD/AOD file They can be used for comparison with truth/reconstructed information It is possible to re-play the trigger decision, by running the hypothesis algorithms on these objects Only the settings of the hypothesis algorithms can be changed in the analysis The effect of different threshold settings can be measured ProductionAnalysisData taking
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 28 Trigger & Physics Weeks Follow up on TDAQ and Computing operation e.g. where are the bottlenecks on (instantaneous and average) rates? Updates on trigger slices and initial feedback on Trigger Aware Analyses in Release 12 Work towards ultimate offline efficiencies and rejections, to be compared to corresponding figures for trigger selections Especially relevant for Combined Performance e, t, m, b-tag Follow up on minimum-bias event selection and analysis Follow up on menu for L = 1 × cm -2 s -1 (items and rates) Small group formed to study the issue and propose a strategy Ideas shown need to be developed into menus for commissioning and for early physics (optimisation for two purposes somewhat different) Follow up on physics with b-jets Efficiency to identify the jets for tagging in LVL2 in various physics scenarios - e.g. rates and RoI multiplicities “Hot topics” End-cap LVL2 muons Forward-jet trigger ETmiss rates (including beam backgrounds) Next week end of October
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 29Planning… Schedula in revisione alla luce delle nuove informazioni su LHC: discussioni durante la TDAQ Week di Settembre: cioè adesso, in perfetta sovrapposizione con la nostra CSN1 (Ho il volo per Londra domani mattina)
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 30 ROS plan Prima tranche dei ROS già pagata al CERN (275 KCHF) Seconda tranche e resto delle network card ordinata e consegna ultimata di recente, pagamento previsto entro la fine dell’anno (275 KCHF) Questo esaurisce il nostro contributo CORE al Read-Out System Impegno importante previsto in ME per contributi all’installazione e commissioning dell’intero sistema Sezioni coinvolte Pavia, Roma1, …
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 31 Data Collection plan Other DataFlow PCs Price enquiry finished for SFI, DFM, L2SV and pROS Evaluations nearly complete Choose which company(ies) to buy from SFOs - specification to be finalised in ~1 month 2007 CORE expected from Italy: 50 KCHF Try and setup system capable of deliverying steady-state nominal SFO output from day 1 Better position to exploit early running trigger and detector studies No correlation with total resources available in HLT farms e.g. send LVL1 output directly to off-line
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 32 Networks and Online plan Switch Procurements (Chassis + Pizza Boxes) Plan finalised, shopping list produced for 2006 Sharing between FA’s agreed File servers and Monitoring PC specs finalised 135 KHCF su CORE 2006, inclusi monitoring PC e local file server: end of contributions
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 33 Infrastructure plan SDX Infrastructure Orders started - cable ladders, nuts & bolts, shelves for switches About to decide on power distribution Coolers for SDX Order placed for 43 racks (all of upper level) 2007 CORE expected from Italy: 80 KCHF
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 34 HLT farms plan HLT Processors IT Market Survey completed Specifications has been drafted: similar (but not identical) to ATLAS needs Not possible to use a blanket contract for 2006: buy our own Lively discussion about spending profile for this items First iteration ended Saturday September 3:44 (Big Ben time) Propose today our best approach based on Need to cover basic needs for commissioning, calibrations, first run period Avoid buying anything that is not absolutely necessary 95 KCHF back to CSN1 in 2006 Other HLT/DAQ Items First bills are appearing here - Rack coolers, inter-rack fibres, etc TDAQ Resource Committee
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 35 Revised Cost Profile (KCHF) Total Pre-series Detector R/O LVL2 Proc Event Builder Event Filter Online Infrastructure INFN Total TDR Total
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 36 Richieste ME (DAQ/HLT) Per le responsabilità vedi la prima trasparenza Installazione e commissioning ROS e HLT 4 mesi uomo a Roma1 2 mesi uomo a Pavia Commissioning DAQ muoni/pixel 4 mesi uomo a Roma1 (incluso ROD Crate DAQ) 2 mesi uomo a Pavia (incluso sviluppo Monitoring) 9 mesi uomo a Genova (incluso HLT steering) Event Building, SFI e SFO 8 mesi uomo a LNF (inclusi Large Scale Tests) Deployment degli algoritmi di selezione on-line 2 mesi uomo a Pavia
CSN1 Trieste Settembre 2006 V. Vercesi - INFN Pavia 37Conclusioni Stato attuale del progetto HLT/DAQ ben allineato con le scadenze future di ATLAS e di LHC previste nel 2007 Il progetto è certamente complesso e anche le responsabilità e i finanziamenti italiani coprono diversi settori Il sistema di HLT/DAQ ha legami profondi con tutte le altre aree di sviluppo: rivelatori, software online e offline, networking, performance di fisica, etc Sono necessari maggiori contributi alla forza lavoro per la parte di installazione e test Impegno fondamentale di consentire a tutti i rivelatori una fase efficiente di commissioning Inserimento degli algoritmi nei test di cosmici Realizzazione di catene complete di read-out e successivamente di Event Building Le slice HLT hanno raggiunto uno stadio di maturità avazata, sono integrate nell’ambiente on-line e costituiscono gli ingredienti per la costruzione e lo studio dei Menu di Trigger Il Progetto è in fase di revisione (management) Necessità di adatttarsi alla nuova fase dell’esperimento Coerenza con l’ATLAS Operational Model Impegno perché gli italiani mantengano i ruoli di visibilità che si sono meritati