ATLAS Referee Report - PowerPoint

Document Sample
ATLAS Referee Report - PowerPoint Powered By Docstoc
					CSN1 Aprile 2006

                        ATLAS HLT/DAQ

Valerio Vercesi on behalf of all people working
    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)
           Pre-series commissioning and exploitation (Everybody)

CSN1 Aprile 2006                      Valerio Vercesi - INFN Pavia                 2
                                                  ATLAS Trigger & DAQ

                                                            Muon        Calo       Inner           Latency    Rates

                                                                                                              40 MHz
        Hardware based (FPGA, ASIC)                                                     Pipeline
        Calo/Muon (coarse granularity)             LVL1                                              2.5 ms

                                                                                       Read-Out               ~100 kHz
                                                            ROD         ROD         ROD Drivers
High Level Trigger

                                                  RoI      ROD
                                                           ROD         ROD
                                                                       ROD         ROD
                Software (specialised algs)
                Uses LVL1 Regions of Interest                                           Subsystems~10 ms
                                                    LVL2     ROB
                                                             ROB        ROB
                                                                        ROB         ROB
                                                                                    ROB hosting
                All sub-dets, full granularity              ROB        ROB         ROB Read-Out
                Emphasis on early rejection                                               Buffers        ~3 kHz

                                                             Event builder cluster
                Offline-like algorithms
                Possibly seeded by LVL2 result                                                        ~1s
                                                                            EF     Event Filter
                Work with full event                                                  farm                    ~200 Hz
                Full calibration/alignment info
                                                           Local Storage: ~ 300 MB/s

    CSN1 Aprile 2006                                Valerio Vercesi - INFN Pavia                                      3
                        TDAQ Networks and Processing
                                                Dual(quad)-CPU nodes
                         ~30                     ~1600      ~100 ~ 500
           Event rate    Local                                                                   Event                               Second-
           ~ 200 Hz     Storage                                                                 Builder    LVL2                      level
  Data                  SubFarm                    Filter                                       SubFarm
                                                   (EF)                                          Inputs     farm                     trigger
 storage                (SFOs)                                                                  (SFIs)
                    DataFlow                                                                                                      stores
                    Manager                                                                                                       LVL2
                                                                                                Network switches                  output

                                                                                                               Gigabit Ethernet
                                                                          Event data requests
                                                   Requested event data

                                                                          Delete commands


  Event data
                          Regions Of Interest

  pulled:                                                                                                                                                                         UX15
  partial events                                USA15                                                                      1600 Data of events accepted
  @ ≤ 100 kHz,                                                                  ~150                                       Read- by first-level trigger
  full events                                                                   PCs                                        Links
                                                                                                                                  VME Dedicated links
  @ ~ 3 kHz                                                                                      Read-Out
                                                                                                                                             Out                       ATLAS
                                                                                                                                           Drivers                     detector
                           RoI                                                                                                             (RODs)     First-
                          Builder                                                                                                                     level
                                                                                                                                                     trigger           UX15
                                                      Timing Trigger Control (TTC)

                                                                                                  Event data pushed @ ≤ 100 kHz,
                                                                                                  1600 fragments of ~ 1 kByte each
CSN1 Aprile 2006                                                                                                                    Valerio Vercesi - INFN Pavia                         4
                          Pre-series system in ATLAS point-1
                                        8 racks (10% of final dataflow, 2% of EF)


   One              RoIB          One        Partial           One            Partial               Partial
   ROS              rack         Full L2    Superv’r          Switch           EFIO      EF rack    ONLINE
   rack               -           rack        rack             rack            rack         -        rack
     -             TC rack          -           -                 -                -      TDAQ          -
  TC rack          + horiz.     TDAQ rack     TDAQ             TDAQ            TDAQ        rack     TDAQ rack
  + horiz.         cooling          -          rack             rack            rack         -           -
  Cooling             -          30 HLT          -                -               -                 4 HLT PC
     -                                                                                   12 HLT
                   50% of         PCs       3 HE PCs          128-port       10 HE PC      PCs     (monitoring)
 12 ROS                                                                       (6 SFI -
                    RoIB                                      GEth for                               2 LE PC
48 ROBINs                                                      L2+EB          2 SFO -                (control)
     underground :                                                            2 DFM)
                                surface: SDX1                                                        2 Central
         USA15                                                                                     FileServers
    •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 Aprile 2006                                    Valerio Vercesi - INFN Pavia                                  5
                   Commissioning and exploitation
   Fully functional, small scale, version of the complete
            Equivalent to a detector‟s „module 0‟
   Purpose and scope of the pre-series system
            Pre-commissioning phase
                  To validate the complete, integrated, HLT/DAQ functionality
                  To validate the infrastructure, needed by HLT/DAQ, at point-1
            Commissioning phase
                  To validate a component (e.g. a ROS) or a deliverable (e.g. a Level-2
                   rack) prior to its installation and commissioning
            TDAQ post-commissioning development system
                  Validate new components (e.g. their functionality when integrated into a
                   fully functional system)
                  Validate new software elements or software releases before moving
                   them to the experiment

CSN1 Aprile 2006                          Valerio Vercesi - INFN Pavia                        6
                                   Pre-series tests at Point 1
   Used integrated software release (                                                                  1-20 L2PU nodes with mufast at point 1 with
    installation image ) with offline release                                                                Muon - 20K Muon events (1-4 L2PU        Factor 1.9
                                                                                                                     applications/node)            improvement
    10.0.6, Event Format version 2.4, TDAQ                                                16000                                                                 respect to one
                                                                                          14000                                                                application/node
    release 01-02-00, HLT release 02-01-01

                                                             Throughput (Hz)
                                                                                                                                                                  1 l2pu/node
   First time e/γ- and μ-selections run in a                                             10000
                                                                                                                                                                  2 l2pus/node
    combined menu with algorithms                                                                        6000
                                                                                                                                                                  3 l2pus/node
                                                                                                                                                                  4 l2pus/node
            muon                                                                                        4000
            calorimeter                                                                                    0
                                                                                                                0          5       10          15        20
            inner detector
                                                                                                                               L2PU nodes
   E.g. Level-2 setup
            8 ROS emulators with preloaded data
            Data with Level-1 simulation: di-jets                                                                      LVL2 Farm Load Balancing
             (17 GeV) , single e (25 GeV), single μ
             (100 GeV)

                                                                               LVL2 Decision Rate (Hz)
            Dataflow applications with                                                                   300
             instrumentation  measure execution                                                          250
             times, network access times and                                                              150
             transferred data sizes                                                                       100
            Used recently up to 20 Level-2                                                                50
             processors each with up to 4
                                                                                                                    1     3    5        7     9     11    13     15     17
                                                                                                                                            CPU Index

CSN1 Aprile 2006                               Valerio Vercesi - INFN Pavia                                                                                                       7
                                              Infrastruttura Event Filter
   Caratteristiche principali SW infrastruttura EF
             Completo disaccoppiamento tra
              data flow (EFD) e data processing (PTs)
                      sicurezza trattamento dei dati                                             Muon     Calo      Inner

             Massimo sfruttamento delle architetture SMP
             Design flessibile e completamente configurabile                              LVL1

                                                                               SFI        RoI     ROD      ROD       ROD
            Node n
                                   Input      EFD
                                                                                           LVL2   ROB      ROB       ROB
             PT P
                                  Sorting                P
                                                         T PT

             #1 I
                                                         I #a
                             ExtPTs      ExtPTs
                                                                                            EF    Event builder network
             PT P                                                                         SubFarm
                T                                        P                                         SFI   SFI   SFI   SFI
             #2 I                                        T PT
                   O                                     I #b
                                  Trash                  O

                          Output Output       Output

                     SFO              SFO               SFO                                        SFO   SFO   SFO   SFO
                   [debug]            [std]            [calib]
                                                                                                   Storage: ~ 300 MB/s
CSN1 Aprile 2006                                           Valerio Vercesi - INFN Pavia                                    8
                                                                                            EF tests
   Verifiche e studi sulla parte infrastrutturale
            Ottimizzazione del protocollo di comunicazione
             tra EF e SFI/SFO:
             miglioramento delle performance per eventi
             piccoli (calibrazione) e farm remote
            Aggiunta di funzionalità addizionali
   Integrazione e validazione degli algoritmi di
            Algoritmi derivati dall'offline
            Ma condizioni operative diverse, es:
                  adattamento delle job-option all'online
                  concorrenza nell'accesso al DB
            Integrata e validata la muon slice
            Altre slice in corso di validazione
   Tested with timing: EF-only, 9 EFDs per 2 PTs,
    TrigMoore algo, 1 MySQL (CERN site)
            All 9 nodes connect to MySQL simultaneously
            all 18 PTs do not 1 but 3 connections to CDI
             (3x18=54 - fast scaling)
                  6.900.20 s   – geometry
                  0.100.03 s   – AMDCsimRecAthena
                  0.060.03 s   – magnetic field
            DB-caching was used
CSN1 Aprile 2006                                             Valerio Vercesi - INFN Pavia              9
                                               HLT Core Software
   Work plan defined for design review 2005
   HLT compliant with trigger operation
            Steering and sequencing of algorithms
            Integration with most recent TDAQ software
            Cycling through TDAQ state machine (start/stop/reinitialize/…)
            HLT trigger configuration from data base
            Use of conditions DB in HLT
            Integration with online services for error reporting and system monitoring
            Many of these issues have a direct impact on selection algorithms 
             Functionality needs to be available early in core software to give time to
             algorithm developers.
   System performance optimization  instrumentation for
    measurement of network transfer times, data volumes and ROS
    access patterns ( complementary to work in PESA group)
   For commissioning and readout tests
            Basic fault tolerance
            Stability
CSN1 Aprile 2006                        Valerio Vercesi - INFN Pavia                      10
                                Software Installation Image
                   Setup / installation scripts                       Originally developed for
                                                                       Large Scale Test 2005
                     Software repositories                            Contains a consistent set of
                                                                       all software in one file
                                                                       needed to run a standalone
                           TDAQ                                        HLT setup
                                                                      Completely tested before
     Common                                       HLT                  deployment by PESA, HLT
                                                                       and DAQ specialists
                                                                      Used for first exploitation of
                           Offline                                     pre-series
                                         Project builds
                                                                      Useful for outside CERN
            Example Partitions / Data Files                            installations and new test
                                                                       bed setups
                           Test suites                 P1 installation procedure
                                                        presently being worked out
                ~ 6.5 GByte software                     Future images snapshot
                                                        of P1 installation
CSN1 Aprile 2006                              Valerio Vercesi - INFN Pavia                              11
                   Trigger Configuration Data Base
     LVL1 + HLT as integrated system                  • GUI for DB population
                                                      • menu changes for experts (HLT and
    offline user    shift crew        expert            LVL1)
                                                      • stores all information to configure the
        TriggerTool              DB population          trigger: LVL1 menu, HLT menu, HLT
                                 scripts                algorithm parameters, HLT release
                                                      • Versions identified with key 
                                                        Configuration and Condition DB
            TriggerDB                  compilers
                                                  Retrieval of information for running:
          R/O interface     Configuration get information via a key, either as:
                            System                • XML/JobOption files
                                                  • direct DB read-out
     online         offline                       for both online + offline running
     running        running

CSN1 Aprile 2006                               Valerio Vercesi - INFN Pavia                            12
                                                        Global Monitoring Scheme

                             Event Monitoring Service

                                                                               Online Histogramming Service
                   ROD                                       Detector



                                                                                                              Monitoring Data

                                                         Event Displays

CSN1 Aprile 2006                                                 Valerio Vercesi - INFN Pavia                                                  13
                                                          GNAM Monitoring
                   Principio: disaccoppiare e mascherare le azioni
                          comuni dagli algoritmi di monitoring
                                                                                        Eventi           Comandi
  GNAM core: azioni comuni                           Dal
          sicronizzazione con la DAQ                                                        GNAM

                                                                     Event Monitoring

          campionamento degli eventi                                                        CORE


          decodifica della parte detector-ind
                                                                                          USER LIB
          pubblicazione e salvataggio degli histo                                         USER LIB
          gestione dei comandi (update, reset, rebin)                                       USER LIB
          tools per gli algoritmi
                  (circular buffer, histogram flags,
                   histogram metadata, ...)
  Algoritmi di monitoring (librerie dinamiche a run-time)                                                Viewer
          decodifica detector-dependent                                                                 Checker
          booking e filling degli istogrammi
          gestione di comandi specifici

CSN1 Aprile 2006                                Valerio Vercesi - INFN Pavia                                              14
             Online Histogram Presenter (OHP)
   Interactive presenter developed in close
    connection to GNAM monitoring
   However used to display histograms                                                               set of
                                                                                                 histograms in
    published on the OHS by any producer                                                              tabs

   Designed to be used both as
            expert mode: a browser of all the                  Browser part
             histograms on OHS
            shifter mode: an histogram presenter to
             show only predefined sets of
             histograms in configured tabs
   Completely interactive with the GNAM
    Core (rebin, reset, …)
   Completely redesigned, after the CTB
    experience, to minimize network traffic
    and to have a scalability appropriate for
    whole ATLAS
            A very useful collaboration with
                                                       Commands to the Core : rebinning, reset
             Computer Science students has been                         ...

CSN1 Aprile 2006                           Valerio Vercesi - INFN Pavia                                          15
                                 Monitoring: commissioning
   Sviluppato un sistema di monitoring/analisi/
    validazione on-line dei rivelatori basato su GNAM
            produzione di istogrammi visualizzati con
             On-line Histogram Presenter (OHP)
            on-line event display (in collaborazione con Saclay)
   In uso al commissioning dal settembre 2005
   In sviluppo
            reperire la configurazione dei rivelatori da DB
            controlli automatici e generazioni di allarmi
   Utilizzato da Tile e MDT, interesse espresso da altri

CSN1 Aprile 2006                                     Valerio Vercesi - INFN Pavia   16
                                                              ROD Crate DAQ
   RCD usato come interafccia verso i RODs per Control,
    Configuration, Monitoring, Data readout (via VME)
   Gli sviluppi RCD hanno avuto sostanzialmente due fasi
            ReadoutApplication (ovvero l'applicazione che costituisce il ROD Crate
             DAQ, il ROS ed il Data Driven Event Builder) modificata in modo
             sostanziale per accomodare tutte le richieste dei rivelatori ed essere pronta
             con tutte le fuzionalità necessarie per il commissioning
                  accesso standardizzato ad Information Service ed Online Histogramming
                  possibilità di accesso ai dati in risposta agli interrupt
                  semplificazione della costruzione delle classi per il controllo e l'acquisizione dei
                  definizione e realizzazione di un data driven event builder
                  librerie per gestione standardizzata delle condizione di errore
            Supporto dei rivelatori per il commissioning
                  Nuovo sviluppo necessario per garantire tramite una semplice interfaccia
                   comune a RAL/CORAL che l'accesso al database di configurazione sia thread
                   safe (fase di inizializzazione)

CSN1 Aprile 2006                              Valerio Vercesi - INFN Pavia                                17
                                                                               Attività RCD
   Parte specifica del detector del ROD Crate DAQ di MDT ed RPC
   Database
            database di cablaggio (molto lavoro!)
            database di configurazione
            Interfacce di online e monitoring con questi
   Detector Control System (DCS)
            Italiana tutta la parte di DCS degli RPC ed il controllo di HV e LV degli MDT
   Settore 13 Muoni
            Run combinati MDT-Tile triggerati da scintillatori
            Studi di sincronizzazione

CSN1 Aprile 2006                                    Valerio Vercesi - INFN Pavia              18
                                                      MDT online calibration
   Required precision for t0 and r-t autocalibration needs inclusive muon rates of 0.33 KHz
            Not suitable for EF calibration streams
            Need different Event Building and streaming (under study)
            Already possible using LVL2 infrastructure with some modifications

                   Thread       L2PU                         Thread   L2PU
                                                                              x 25
            Thread            Thread            Thread

                                       TCP/IP, UDP,                               Local            Local
                                                         ~ 480 kB/s              Server           Server
                                                                                          x ~20

                                       Server                                                     ~ 9.6 MB/s

                            Memory                                    Gatherer                    Calibration
                             queue          Dequeue                                                  farm
                                                             ~ 480 kB/s
CSN1 Aprile 2006                                        Valerio Vercesi - INFN Pavia                            19
                        SDX1 – TDAQ Room @ P1

Total of 99 racks can be placed in SDX
    •Lower Level: 49 (LVL2, EB,…)
    •Upper Level: 50 (EF)

CSN1 Aprile 2006                 Valerio Vercesi - INFN Pavia   20
                                                                                                                                                   ROS Overview
          SDX1                       dual-CPU nodes
              ~30                    ~1600                                          ~100 ~ 500                              • In total ~150 ROS PCs will have to be
   Event rate Local                                                                 Event
   ~ 200 Hz Storage
                                                                                               LVL2                           installed
                                       (EF)                                           Inputs
                                                                                                                            • Each ROS PC will be equipped with 3 or
                                                                                                                pROS          4 ROBIN cards and one 4-port G-bit
                                      switches                                                                                Ethernet NIC
                                                                                                                                                       ROBIN    ROS PCs in USA15
                                                                                    Network switches
                                                              Event data requests

                                       Requested event data

                                                              Delete commands

                                                                                                   10-Gigabit Ethernet
               Regions Of Interest

                                     USA15                                                                               Data of events accepted
                                                                                                                   1600 by first-level trigger
                                                                    ~150                                           Read-
                                                                    PCs                                            Out    VME Dedicated links
                                                                                                                           Out                       ATLAS
                                                                                                                         Drivers                     detector
                  RoI                                                                                                    (RODs)        First-
                 Builder                                                                                                               level
                                                                                                                                      trigger       UX15
                                          Timing Trigger Control (TTC)

CSN1 Aprile 2006                                                                                                            Valerio Vercesi - INFN Pavia                           21
                          ROS Hardware Procurement
    ROS PCs
  1st batch (50 PCs)                           Ordered and received

  2nd batch (60 PCs)                           Ordered. Delivery scheduled for May

  Remaining ROS PCs + spares                   Will be ordered later

   German production (350 cards)   Ordered and received (~20 cards did not pass the production
                                   test and still need to be repaired)

   UK production (350 cards)       Ordered. Delivery scheduled for March

     4-port NICs

      Ordered. Delivery scheduled for May                           Silicom
                                                                    4-port NIC
CSN1 Aprile 2006                          Valerio Vercesi - INFN Pavia                           22
             Current Status of ROS-Racks in USA15
                                               Liquid Argon                                     TileCal
                    Y.09-16.A2   Y.08-16.A2   Y.07-16.A2 Y.06-16.A2   Y.05-16.A2   Y.04-16.A2   Y.09-14.A1
   Control switch
        ROS PCs


       Power &
       network cables   yes        yes           yes        yes           yes          yes         yes
       (ROS level)      yes        yes           yes        yes           yes          yes         yes
       (ROD - ROS)
                        50%        no            no         no            no           no          50 %

CSN1 Aprile 2006                                   Valerio Vercesi - INFN Pavia                              23
                   Physics and Event Selection Architecture
   PESA Core SW is responsible for the implementation of the
    Steering and Control (built around standard Athena components)
   PESA Algorithms develops HLT software using realistic data
    access and handling
            specialized LVL2 and EventFilter algorithms adapted from on-line
            deployment in HLT testbeds
   PESA Validation and Performance evaluates algorithms on data
    samples to extract efficiency, rates, rejection factors, and physics
   Stems from original structure, laid out in parallel with the
    organization of the Combined Performance working groups, in
    “vertical slices" (LVL1+LVL2+EF)
            Electrons and photons
            Muons
            Jets / Taus / ETmiss
            b-jet tagging
            B-physics

CSN1 Aprile 2006                       Valerio Vercesi - INFN Pavia             24
                   HLT Reconstruction Algorithms
   HLT Feature extraction algorithms are available for each slice
   Calorimeter algorithms
            LVL2 and EF algorithms ready for e/g
            t implementation ready at LVL2
            Offline tool adapted to the EF is ready for JetCone
   Muon algorithms
            LVL2 and EF algorithms are available for the barrel region; work has
             started on extending the LVL2 algorithm to the endcap
            ID to muon track matching tools are available at LVL2 and EF
            Muon isolation studies using calorimeters are being performed
   ID tracking
            Tracking with Si data ready at LVL2 and EF; more approaches studied in
            Tools available for both track extension to the TRT and stand-alone TRT
             reconstruction; emphasis on providing a robust tool for commissioning and
             early running

CSN1 Aprile 2006                         Valerio Vercesi - INFN Pavia                    25
                             Selections: LVL2 m
                            Implemented curvature radius instead of sagitta
                                  More suitable for the endcap, recover efficiency in
                                   the barrel
                                  Same algorithm across ± 2.4 in h

                          Resolution
                                   New LUTS for Radius
                                   Slightly worse than 10.0.3
                                   Resolution is OK for Standard sectors
                          Turn-on curves
                                   11.0.3 comparable with 10.0.3
                                   Resolution is OK for Standard sectors
                                   Worse efficiency in the feet region (Special

                          Endcap extension in progress
                          Combined reconstruction (mComb) with ID
                                   Refine the mFast pT by means of ID data
                                   sharper 6 GeV threshold

CSN1 Aprile 2006   Valerio Vercesi - INFN Pavia                                          26
                                                         LVL2 cosmics m
                   X-Y                                     Z-R: bending plane


    Straight line
    from y=+98.3 m              MDT hits are
                                station centers
                                in X-Y.

            MDT hits                              /castor/
            RPC hits (pair of phi,eta strips)     Monte Carlo!
            Muon track from the surface
      MDT,RPC hits are there and looks fine. Conversion of RDO to
      coordinates seems fine too. Next steps: MuFast modifications

CSN1 Aprile 2006                         Valerio Vercesi - INFN Pavia                        27
                                                          Selections: EF m
   Studies on single muon selections have been performed for two scenarios:
            6 GeV threshold at 1033cm−2s−1 luminosity and 20 GeV at 1034cm−2s−1.
   Cuts are defined so that a 95% efficiency is achieved at the threshold values.

                                                      Sorgenti di     L=1034       L=1034         L=1034
                                                        muoni        No backgr.    s.f. x1        s.f. x5

                                                         /K              54 Hz      54 Hz         48 Hz

                                                          b               77 Hz      77 Hz         68 Hz

                                                          c               30 Hz      30 Hz         26 Hz

                                                          W               22 Hz      22 Hz         19 Hz

                                                           t         negligible    negligible    negligible

                                                        Total         ~185 Hz       ~190 Hz       ~180 Hz

                                                      Layout Q (barrel only)
     lower values of efficiency plateau              MuId Combined used at EF
     less sharp curves near the thresholds           MuComb rate reduction still to be included at LVL2
     more points are needed for a better curve       Fake rates expected to be ~1% (~12%) of total rate
      definition                                       for s.f.x1 (s.f. x5) with this threshold (seeded mode)

CSN1 Aprile 2006                           Valerio Vercesi - INFN Pavia                                         28
                                                   Selections: b-tagging
   Two classes of tagging variables can be used: track variables (xT ) and collective
    (vertex) variables (xV ).The weight of each RoI is computed using the likelihood-
    ratio method

      where Ssig and Sbkg are the probability densities for signal (b-jets) and
            WT : transverse (d0/sd0 ) and longitudinal (z0)
            WV : secondary vertex energy and mass (statistical approach)
   Recent work to combine SimpleVertex (1-dim fit) and VKalVrt (offline algorithm adapted
    to LVL2)

                                                              Impact parameters
                                                              Impact parameters + probabilistic vertex
                                                              Impact parameters + VKalVrt/SimpleVertex combined
CSN1 Aprile 2006                                   Valerio Vercesi - INFN Pavia                                   29
                                  Trigger-aware analysis
                                                                       Analyses using trigger
                                                                        information as a “pre-
              Data taking Production     Analysis                       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
                                                                       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

CSN1 Aprile 2006                       Valerio Vercesi - INFN Pavia                                  30
                   Trigger & Physics Weeks
                               Aim: bring together trigger, detector
                                performance and physics studies and expose
                                trigger issues and strategy to a broad ATLAS
                               Focus on initial scenario: 1031cm-2s-1, 200Hz

CSN1 Aprile 2006      Valerio Vercesi - INFN Pavia                              31
                                                                        Ideas @ 1031
   Electrons / Photons
            Sketch some pre-scale factors @ HLT                             HLT       Rate (Hz)
            Crude estimates “to guide the eye” keeping
             total e/g output rate constant                                   e10i        2           250
            Photons not yet worked out                                       e15i        5            40
                  Assessment of both di-photon thresholds and
                   high-pT single one to be revisited                         e20i        36           1
                  Photons useful to obtain unbiased jet sample              2e10i       ~Hz           1
   Muons
            Absent (or very low) cavern background
                                                                                           Rate      Pre-scale
             makes LVL1 commissioning “easier”                                             (Hz)      factor
                  Full shielding, 75 ns bunch spacing
            Build menus allowing to                                         LVL1 m6           40       6
                  Measure Xsec from (very)-low pT                           HLT m6            20       3
                       Can go as low as ~5 GeV
                  Check W, Z, J/, Y…                                   LVL1 m20              14       1
                  Study ways to increase trigger acceptance             LVL1 dim              3        1

CSN1 Aprile 2006                              Valerio Vercesi - INFN Pavia                                       32
   Contributo INFN alla Pre-serie
            140 KCHF (ROS Racks, Monitoring, Operations, Switches, FileServer)
             completamenti spesi entro il 2005
            Per questo e per il resto VV riceve in copia tutte le fatture
   Contributo CORE 2005-2006
            Online Computing System: 45+135 KCHF (Monitoring, Operations)
                  Inviati al CERN 45 KCHF a Maggio 2005
                  Già acquistati quattro file server
            Read-Out System: 275+275 KCHF (ROS Racks)
                  Gara CERN espletata con un congruo ritardo per la prima tranche (50 ROS), la parte
                   rimanente è in consegna (60 ROS a Maggio)
                  Imputati all‟INFN per ora circa 200 KCHF (su Roma1)
            LVL2 processors, Event Building, Event Filter processors:
             65+50+170 KCHF
                  In corso di perfezionamento le specifiche dettagliate (soprattutto per i processori HLT)
                  Può darsi si possa utilizzare un marker survey fatto da CERN-IT
                  Studi in corso anche per la valutazione delle ultime tecnologie (Moore‟s law failures…)
            Infrastruttura: 80 KCHF (cavi, racks, cooling,…)

CSN1 Aprile 2006                                Valerio Vercesi - INFN Pavia                                  33
                                       Cost Profile (KCHF)
                         2004   2005       2006        2007     2008   2009   Total

  Pre-series             140     0           0           0       0      0      140

  Detector R/O            0     275        275           0       0      0      550

  LVL2 Proc               0      0          65          195     230    160     650

  Event Builder           0      0          50          50      110     70     280

  Event Filter            0      0         170          180     570    380    1300

  Online                  0      45        135           0       0      0      180

  Infrastructure          0      0          80          80       20     20     200

          INFN Total     140    320        775          505     930    630    3300

           TDR Total     1048   3357       4087        4544     7522   4543   25101
                         13.4   9.5        19.0        11.1     12.4   13.9   13.1
    INFN Percentage(%)
CSN1 Aprile 2006                 Valerio Vercesi - INFN Pavia                         34
                                                               INFN Milestones
   30/06/2005
            TDAQ - Installazione, test e uso della "Pre-serie"
             (~ 10% TDAQ slice)
                  “compiutamente” raggiunta in Ottobre: ritardi accumulati soprattutto sugli acquisti delle
                  Proposta di indicare il 100% e modificare la “matching date”
   24/12/2005
            TDAQ - Installazione e test dei ROS di Pixel, LAr, Tile, Muon (interfacciamento al
             ROD Crate e integrazione nel DAQ)
                  Forte dipendenza dalla data di consegna dei ROS (lentezza gara, etc)
                  Nessun problema “di principio”, il programma di lavoro è chiaro, l‟esperienza della pre-
                   serie è direttamente trasferibile
                  Proposta di indicare 50% alla data prevista
   30/04/2006
            Completamento dei test sulla pre-serie e definizione delle funzionalità per il
             supporto al commissioning TDAQ
   31/08/2006
            Commissioning delle slice di ROS dei rivelatori utilizzando le funzionalità della pre-
             serie (modulo-0 del sistema finale)
   31/12/2006
            Presa dati integrata dei rivelatori nel pozzo con raggi cosmici

CSN1 Aprile 2006                                Valerio Vercesi - INFN Pavia                                   35
   Il progetto TDAQ sta entrando in una fase di piena
            Rendere disponibile ai rivelatori tutte le infrastrutture
             necessarie per i run di cosmici
            Preparare il commissioning completo del sistema in
             preparazione allo start-up di LHC
   I contributi italiani sono chiaramente visibili e ben
    riconosciuti a livello della Collaborazione
            Integrazione hardware, sviluppi algoritmici, posizioni di
             responsabilità, finanze
   Il tempo a disposizione per il commissioning TDAQ è
    molto compresso
            Fondamentale per poter assicurare il data flow necessario
             anche allo start-up
CSN1 Aprile 2006                     Valerio Vercesi - INFN Pavia                 36
                    Goal of Early Commissioning…

                   Prepare for Unexpected Events…
CSN1 Aprile 2006               Valerio Vercesi - INFN Pavia   37
                                                                                                                 LVL2 tests
                                               LVL2                                                                                  #Reqs    Data
  Data File                  LVL2 Rate         Latency      Processing Time        RoI Col   DAQ Time        Data Rate   Data Size   /Event   /Req
                                    (Hz)         (ms)              (ms)              (ms)      Fraction        (MB/s)      bytes              bytes
  mu                               293.1         3.41              2.78              0.62        0.19           0.084       287        1.3     223
  jet                              280.3         3.57              3.26              0.28        0.09           0.781       2785       1.2     2283
  e                                 58.2         17.18            15.48              1.66        0.10           0.921      15820       7.4     2147

                                         Fraction of events passing LVL2 as a function of
                                                       the decision latency

              Fraction of events

                                   0.8                                                                                                  mu
                                   0.6                                                                                                  jet
                                                                                               Prefiltered                              e
                                           1   3        5   7     9   11 13 15 17 19 21 23 25 27 29
                                                                              Latency (ms)

CSN1 Aprile 2006                                                              Valerio Vercesi - INFN Pavia                                            39
                      Example: Data Base Schema
                                                       keys: stored in CondDB, to
                                                       retrieve information (online and

    LVL1                                                                         HLT

                   Early prototype of HLT part
                   already run on 6 node system
                   with muon selection algorithm
                                                               algorithms,   software
                                                  trigger menu jobOptions    release

CSN1 Aprile 2006                    Valerio Vercesi - INFN Pavia                          40
                   Routing m calibration data

CSN1 Aprile 2006       Valerio Vercesi - INFN Pavia   41
                                                                   Selections: e/g
    Rate and efficiency studies performed for                                   Eff %      Rate
     main physics triggers: e25i , 2e15i, e60, g60,                 L1           95.5      4.7 KHz
    Results for 11.0.4 perfectly in agreement with             L2 Calo          94.9      890 Hz
     Rome results                                                  L2 ID         91.0      280 Hz
    Tools have been developed to optimize the
     selections                                                L2 Match          89.7       98 Hz
    In the future, results will be provided as                 EF Calo          87.6       65 Hz
     efficiency vs. rejection curves, to provide a                 EF ID         81.8       35 Hz
     continuous set of working points: essential for
     trigger bandwidth optimization                            EF Match          81.0       35 Hz

                                                              Cluster composition
                                                            We                              21%
                                                            Zee                              5%
                                                            Direct photons or quark brem      5%
                                                            e from b, c decays                37%
                                                            rest                              32%

CSN1 Aprile 2006                              Valerio Vercesi - INFN Pavia                           42
   LVL2 calo algorithm for taus recently separated from egamma
   Ongoing performance studies for selection strategies on variables
   At present only EM calibration for cluster energies: need for a tau
    calibration (also for EF, H1 style as in the offline mode?)
   First implementation of EF “seeded” TrigtauRec is already working
    making use of offline tools
   Once the selection strategies are defined, physics trigger-aware
    analyses (studying the effect of the hadronic tau trigger) can be
   Three different strategies (concerning the data preparation) are
    being considered
            Read out calorimeter and unpack the cells (unpacking time may dominate)
            Read out calorimeter, get Ex/ Ey calculated in ROD (faster but …
            Read out TriggerTower from LVL1 Preprocessor
   Ongoing work to define and studies general strategy for pre-scales,
    in particular for jet objects
CSN1 Aprile 2006                       Valerio Vercesi - INFN Pavia                    43
                   Jet triggers and prescales

CSN1 Aprile 2006       Valerio Vercesi - INFN Pavia   44
                                              RoI Based B-physics
   Aim: use the calorimeter to identify                  LVL1 EM RoI multiplicity vs. ET cut
    regions of the event containing B
    decay products
            EM RoI for e and gamma.
            Jet RoI for hadronic B-decays                                         Towerthresh=500MeV (default)

   Keep multiplicity low, to minimize                                             Towerthresh=750MeV

                                                   RoI Multiplicity
    data transfer and cpu, whilst
    maximising efficiency for events                                               Towerthresh=1000MeV
    used in physics studies
            multiplicity= 1-2
   The effect of different thresholds
    (EM&HAD and the jet RoI size on
    this multiplicity was studied using
    Rome data (1x1033) with the new
    TTL LVL1 simulation and pile up
   The requirement on multiplicity
    implies an ET threshold of ~ 2GeV                                 LVL1 Threshold Energy (GeV)
    for LVL1 EM RoI

CSN1 Aprile 2006                             Valerio Vercesi - INFN Pavia                                     45
                                      2006 PESA Milestones
   LVL1/HLT AODs fully available in Rel 12 for                              T&P Week
    trigger-aware analyses – Apr 06
            Very preliminary AOD information available in Rel 11
            Detailed description of Rel 12 deliverables prepared by Simon
   HLT algorithm reviews complete – Jun 06                                  T&P Week
            Detailed review of ID LVL2 algorithms already taken place
            Focus on system performance and implementation
            Results fed back into Rel 13
   Online tests of selection slices with preloaded mixed files and large
    menus – Sep 06
            First production version of trigger configuration
   Selection software ready for cosmic run – Oct 06
            Already in PPT: need to refine meaning
   Blind test of HLT selection – Dec 06                                     T&P Week
            In discussion with physics coordination
            Sample of representative events from initial ATLAS output & run full menu

CSN1 Aprile 2006                         Valerio Vercesi - INFN Pavia                    46
                                                                                                                  PESA Planning
   Several interactions with PESA Slice coordinators and with Algorithms developers
   Try and bring together something to help reinforcing the content of proposed milestones and
    monitoring the development process
                       Only gone through first iteration until now…
                       Try always to describe the work in a “task oriented” fashion, to help identifying weak areas as well as
                        facilitate the job assignment
   Attempt to build a full PESA planning (Excel) starting from this information to monitor progress and
    allow for updates, suggestions, improvements
                       Clearly more details on near-future objectives than on far-away ones
  PESA Planning
  Task                                                                    Comments                                                                       Expected        PPT Workpackage

  LVL1 Trigger

  Definition of EDM                                                       Done?                                                                          dec-05

  ……………………………………………………………………………………………                                     ……………………………………………………………………………………                                               ………………………       …………………………..


  e/gamma implementation in common framework                              RTT, ESD, Root Analysis Framework                                              February 2006   DH-W101

  Develop tools for automatic optimisations of e/gamma selections         scanning of parameter space, minuit fitting there, neural net, multi-variant   March 2006
                                                                          method being developed                                                                         DH-W101

  Check trigger selection w.r.t offline selection for electrons/photons   Need new evaluations from offline groups                                       March 2006      DH-W101

  Establish set of pre-scaled e-triggers using Rome datasets              Photons as well                                                                February 2006   DH-W101

  First evaluation of trigger efficiencies from data                      For electrons, photons and muons                                               March 2006      DH-W101

  Strategies for ETmiss calculations                                                                                                                     March 2006

  Revised Steering Configuration                                                                                                                                         DH-W110

  Prototype LVL2 Hypothesis algorithm for all                             Examples to be further developed in validation                                 February 2006   DH-W147

  Provide documentation and examples to physics community                 For all selections                                                             March 2006

  Milestone April 2006                                                    LVL1/HLT AODs completely available in version 12 for trigger-aware analyses

CSN1 Aprile 2006                                                               Valerio Vercesi - INFN Pavia                                                                                47

Shared By: