Document Sample
Wu Powered By Docstoc
					Introduction to Geneva ATLAS High
        Level Trigger Activities
                         Xin Wu

   Journée de réflexion du DPNC, 11 septembre, 2007

   Assitant(e)s: Gauthier Alexandre, Francesca Bucci,
                 Till Eifert, Clemencia Mora
   MA:           Olivier Gaumer, Andrew Hamilton,
                 Phillip Urquijo (20/09/07)
   Physiciens: Szymon Gadomski, Xin Wu                  1
The Challenge of Trigger at LHC
                                Bunch crossing              40 MHz
                                σ total                     70 mb
             Event rate 
                                Event rate                  ~1 GHz
                                Number of event/BC          ~25
                                Number of part./event       ~1500
                Level-1        Event size                  ~1.5MB
                                Mass storage rate           ~200Hz
               Level-2 

         Mass Storage 

       Offline Analyses

                             Need to have Trigger of high performance
                                ~6 order of rate reduction
                                Complex event and 140 M channels
    Brief Introduction to the ATLAS Trigger System
                                                                         MuTrigDet Other detectors
  LVL1: Hardware Trigger                                        40 MHz
                                                                                              1 PB/s
  EM, TAU, JET calo. clusters                                   Pipelines
                                      LVL1              2.5 s    2.5 s
  µ trigger chambers tracks          Calorimeter      Muon
                                        Trigger       Trigger    LVL1 Acc.
  Total and missing energy
                                                                 100 kHz     ROD   ROD       ROD

  HLT: PC farms                          RoI’s (Region of Interest)                       120 GB/s

 LVL2: special fast algorithms         LVL2       ~40ms     RoI
                                                                             ROB    ROB      ROB
    Access data directly from             ROIB     L2SV

      the ROS system                                                         ROS
                                     H L2P                RoI data
    Partial reconstruction               L2P        L2N
                                                                                          3 GB/s
      seeded with L1 Regions of L                         LVL2 Acc.
                                                                             Event Builder
      Interest (RoIs)                                     3 kHz
                                     T Event Filter
 EF: offline reco. algorithms                        ~4s
    Access to fully built event             EFP
                                                            EF Acc.
    Seeded with LVL2 objects
                                                          200 Hz
      (full event reconst. possible)
    Up to date calibrations           Event Size ~1.5 MB                               3
                                                                                   300 MB/s
Geneva’s Participation in High Level Trigger
 Calorimeter Trigger Software (Gauthier, Olivier, Xin)
    Overall coordination
    LVL2 calorimeter cluster correction
 HLT Steering Controller (Till)
    Control the complex algorithm scheduling for ROI based
     reconstruction and Stepwise processing for early rejection
     (see Till’s talk)
 Online integration of the HLT algorithms (Xin)
    Integrate the HLT algorithms developed offline into the DAQ
     online running environment
 Trigger Event Data Model (Andrew, Francesca)
    Manage trigger objects stored in data (see Andrew’s talk)
 EF tracking software (Andrew, Francesca)
    Adapt offline track reconstruction for EF (see Andrew’s talk)
 Express stream (Syzmon)
    Special data stream for fast reconstruction
 ATLAS Trigger Coordination (Xin)                           4
Calorimeter Trigger Software
 Collaborative effort of many people
    Common first steps for all the “slices”: electron, photon,
     jet, tau, missing energy
 LVL1 hardware simulation
 Calorimeter RegionSelector
    Mapping between detector elements and - region for
     using Region of Interest
 Calorimeter data preparation
    Fast raw data unpacking
 LVL2 calorimeter reconstruction
    Specific fast clustering algorithms
 LVL2 cluster calibration
    Energy correction, position correction, crack correction,…
 Event Filter calorimeter reconstruction
    Adapt offline algorithms for EF
 Overall coordination                                        5
L2 EM Cluster Corrections (Olivier, Gauthier)
 Lateral energy correction
    Better Energy evaluation (10% effect)
 S-shape correction (sampling 2)
    Better position reconstruction
 Longitudinal energy correction : Material and leakage
    Better energy resolution
 Energy  correction and  correction + accordion modulations
  for different clusters
 Crack corrections (local correction)
     = 0.8 : crack between the two electrodes of the barrel
     = 1.4 : crack between barrel and end-cap
 Currently first 2 corrections implemented using offline
    Study effect on trigger in progress

                                                             From Olivier

          Energy correction - Effects
    Energy calibration based on offline                     Used to give the best
     calibration:                                             energy resolution  Get
                                                              the best efficiency
         global factor (lateral leakage)                   On set of parameters per 
        off : offset                                         position
        wi: weights on pre-sampler and layer 3 energy

    MZ reconstructed from electron pairs
        - With energy correction
        - Without energy correction

                                                                            From Olivier

           S-shape correction study
Function proposed for this correction :                      corr    f (u )   Where   1  u  1
                                                    With                       
                                                            f (u )   0 arctan( 1u )   2 u   3 u   4
This function is actually modified to ensure the continuity at |u|=1
The variables are redefined to remove correlations between them
At the end the actual function used is :
                                                       
             u arctan( )  arctan( u )               
f (u )  0            1            1
                                            2 (1  u )                0.025<<0.05
             Z arctan( )  arctan(Z )                 
             1          1                             
                                                       
Z                  1
      arctan(1 )
                                                                                     . Before correction
 Only 3 parameters left tabulated as
 function of energy                                                                 . After correction
 An interpolation in energy is done
 on the parameters
Online Integration of HLT Algorithms
 Integrate the HLT algorithms developed offline into the DAQ online
  running environment
 HLT algorithms developed in the offline framework because they use
  many offline reconstruction tools (more on EF, less on LVL2)
    Read MC pool RDO files and use transient BS
    Run together with Reconstruction
    Well suited (fast turn-around) for trigger performance studies
 Online running is quite different from offline
    Transition controlled by DataFlow software rather than Athena
    Read ByteStream raw data from ROS through DAQ
    Need to interface to online monitoring/error reporting tools
    Need to be thread-safe for multithreaded running
 Online integration involves many components of the HLT:
    Algorithms, trigger configuration, database, Steering Controller,
      Data Collection, …
    Follow through integration steps from offline, quasi-online
      (Athena MT/PT) tests all the way up till final online validation at
      point-1                                                           9
    Steps of Online Integration

       Offline             Simulated Online       DAQ Data Flow
     Environment             Environment

          athena            athenaMT/PT               L2PU

       Steering                Steering              Steering
      Controller              Controller            Controller

      Algorithms             Algorithms            Algorithms

1) Testoffline         2) Test with athenaMT  3) Test at Point 1
    – RDO input            – simulate online      – actual DAQ
    – Raw (BS) input       – BS input             – BS input
                           – use TDAQ release       (through ROS)

DAQ/HLT Technical Runs
 Dedicated Technical Runs (1 week each) are used to test DAQ/HLT
  and HLT algorithm integration
    So far two in 2007 (March and May). Next in end of September
 Brief Summary of the May TR (21/5-25/5)
    ‘Final’ Hardware
        • ROIB (+ LVL1 emulator), 120 ROSs
        • 4 HLT racks (130 dual quad-core 1.8 GHz), ~5% final system
    tdaq-01-07-00, AtlasHLT 2.0.5-HLT, Offline 12.0.5-HLT-1
    All basic HLT slices integrated
        • e10, g10, mu6, tau10, jet20, cosmic, Bphysics, met
        • combined : e10+g10+mu6+tau10+jet20
    ~ 6k events (mixed physics processes, ~60% jets and ~40% W/Z)
 Main achievement :
    Validated TDAQ and HLT infrastructure with final hardware
    Measurements with dummy algorithm LVL2 and EF with final
    Functionality test with combined algorithm
    Tested DBProxy and triggerDB configuration
 Next Technical Run: Sept 24-30
LVL2 Timing for Rejected Events

   Total time per event           Processing time per event
    mean = 31.5 ms                 mean = 25.7 ms

Data collection time per event   Data requests per event
mean = 6.0 ms                        mean = 5.3

Express Stream (Szymon)
 ATLAS data streams

Calibration streams contain incomplete events.
Complete physics events used for calibration are in the Express.
                                      From Szymon

 Express Stream of ATLAS data
What is the Express Stream
• One of the data streams produced by ATLAS online,
  O(10%) of the physics data.
• To be reconstructed and looked at rapidly. Results in a
  few hours, before the reconstruction starts.
• Calibration, check of data quality, monitoring of the
  detector status, rapid alert on interesting events…
Role of Geneva
• S.Gadomski coordinates the work on the trigger menu.
• Trigger rates are calculated on Swiss ATLAS Grid
  resources, in collaboration with Bern (Sigve Haug).

 ATLAS HLT project is in good progress
    Trigger algorithm development in advanced stage
    Trigger menu for early data-taking being completed
    HLT being integrated online and performance being
     studied in Technical Runs
 Over the pas year Geneva expanded its effort in the ATLAS
  High Level Trigger and made many important contributions
 We are becoming key players in several areas
    Calorimeter Trigger Software, Steering, EDM, Online
     Integration, Express Stream, Trigger Coordination
    See Till and Andrew talks for some more details
 Expertise in HLT is a great advantage for the group to access
  and understand real data at the earliest stage

LVL2 Egamma Reconstruction Algorithm

                4 Processing steps of T2CaloEgamma
                at each step data request is made and
                   accept/reject decision is possible

                                   Rcore= E3x7/E7X7
                                  in EM Sampling 2

                                  in EM Sampling 1

       p0            g
                               EtEm=Total EM Energy
                               (add sampling 0 and 3)

                               EtHad=Hadronic Energy
                                   (Tile or HEC)

Calorimeter Timing Results from the May TR

  T2CaloEgamma               TrigCaloCellMaker
                              mean 16ms / RoI
          mean 6.2ms / RoI

  mean 27ms / RoI
                                     mean 65ms /RoI


Shared By: