Simulation by ICUlV7NM

VIEWS: 38 PAGES: 20

									                   STT simulations
                          (Horst Wahl, 25 February 2000)




    Outline:

    trigger simulation
        (Silvia Tentindo-Repond, Sailesh Chopra, John Hobbs, with help
         from Brian Connolly, Harrison Prosper, + Dave Toback)




    queueing studies
        (Stephan Linn)
               STT trigger simulation
             (Silvia Tentindo-Repond, Sailesh Chopra,…...)

   STT trigger simulator:
       integral component of D0 RunII Trigger Simulator;
       tasks:
           simulate all components of the STT:

                   hit clustering (STC)
                   CTT road information handling (FRC)
                   hit filtering (STC)
                   tracking (TFC)
            provide tool for optimization of
                   trigger parameters and algorithms
                   monitoring parameters
            provide tool for determination of efficiencies

       L2STT Simulator code split into three packages:
          tsim_l2stt (main package)

          l2stt_util

          l2stt_fitting

                 (for detailed studies of track fitting
                 algorithms)
                   STT simulation status
   presently available functionality
       SMT hit clustering:
          cluster algorithm:

                  so far, only one algorithm implemented:
                      neighbor clustering without cap on clustersize,
                   (same alg. as used in L3, but different from
                   algorithm being studied for firmware
                   implementation)
                  other algorithms foreseen
            hit filter:
                  CTT roads now integrated into main package;
                  temporarily, use analytic expression for translation
                   from CFT to SMT coordinates
                  first version of LUT (translation map) available (but
                   based on nominal position -- will need possibility to
                   use real positions)
            tracking (with John Hobbs, Wendy Taylor):
                  l2stt_fitting contains all possible tracking algorithms
                   (for testing);
                  main package will only have final choice of algorithm
                  still being debugged
            test output:
                  create a text file that contains SMT hits in cable-
                   format, to be used as an input for debugging and
                   testing of VHDL code for clustering. (RCP switch)
                  CTT roads: to be done
       most of the code committed to CVS t70
        Clusters from t tbar events
   only from barrels
     Clusters from Zb bbar + 2 min.bias
                   events
   barrel only
       Clusters vs SMT layer
 from Zb bbar + 2 min.bias events




layer   mean    rms
  1     148.6   42.6
  2     113.7   34.0
  3     165.3   48.5
  4     129.0   36.6
  5     112.6   34.9
  6     96.1    29.0
  7     142.6   39.8
  8     129.7   37.9
   Clusters from
    one t tbar
    event




   Clusters from
    100 t tbar
    events
   one t tbar event

       road centers in
        SMT




       SMT clusters
        in CTT roads
CTT tracks from Zb bbar + 2 min.bias
               events




                                 Clusters
                                  in roads




                                 nb. of
                                  CTT
                                  tracks
                    Queueing Studies
                             (Stephan Linn)

   Questions to be answered by queueing studies:
        how much processing (e.g. track fitting) time can we
         afford before deadtime becomes unacceptably high?
        where are the potential bottlenecks in the data flow
         through the trigger?
        additional buffering needed?


   Queueing simulation software
        previously, used RESQ (IBM product)
                  not supported anymore, only runs on IBM platforms
                      look for alternative


   Ptolemy:
        simulation package developed by EE and comp.science
         dept. at UC-Berkeley
        can simulate complex systems at different levels of detail
         and with different time scales
        elements of system represented by “queue and serve
         galaxies”
                  specified by service time and queue depth
                  event defined by starting time and data value (event
                   number)
               STT Model in Ptolemy
   use available specifications
           (note by U.Heintz at
           http://physics.bu.edu/~heintz/STT_q.pdf)
        parameterize (from MC data)
                  Nt = number of tracks per sextant
                  Nh = number of hits per detector
                  H = clusters per hit
                  T = clusters per track
        transmission speeds
        data sizes
   model:
        STT modeled as 6 independent sectors
        random numbers Nt, (hyperexponential) Nh (double
         gaussian), track fitting time (double exponential, from
         JH+WT studies)
        correlations between module delay times (depend on
                               Nt, Nh)
        PCI bus “arbitration” (priority to road transfer over
         filtered clusters)
        STC filter waits for roads from FRC
        (see draft of note by Stephan Linn at
           http://www-d0.fnal.gov/~linn/d0_private/queue.ps)
STT Queueing model
PCI bus model
distributions used

                    cluster
                     multiplicity




                    track
                             multiplicit
                     y




                    track fitting
                     time
               Queuing simulation results

   latency
           for one STT sextant:




            latency for one sextant reproduces track fit delay
            full system latency: worst-of-N convolution with
             five other sextants
            total deadtime: < 1% (for nominal values)
                       (not counting deadtime due to min. time
                       between L1 accept)
            16 event buffer before track fitting takes care of
             50s track fitting time -- never filled to capacity.
            no additional buffer necessary
     Parameters in queueing simulation
   Parameters used in simulation:
       number of hits per detector Nh
          double gaussian with mean = 36, max. = 90
          (corresponds to 8 overlapped jet-brew events,
          with L1 trigger pt > 7 GeV, plus 2% add’l occupancy
          for noise)
       number of L1CTT tracks per sextant Nt
        hyper-exponential with mean = 2 , rms = 4, max = 32
          (corresponds to 6 overlapped jet-brew events,
          with L1 trigger pt > 7 GeV)
       number of hits per cluster H:
          gaussian with mean 3.6, rms 2.8 (from t tbar events)
       timing for tracking: double exponential, with “old”
        values, i.e. 15s
   notes:
       changing Nt to 4.8 with rms 8, other parametes
        unchanged: deadtime < 2%
       Z  b bbar + 2 MB events:
           Nt = 1.5 (9 per event with rms 5),

           Nh = 9, H = 3.9

       for 8 overlapped jet-brew events, Nt = 3.7, rms 6.0
       time for track fitting now much shorter than that used
       for 396ns bunch spacing:
           for L = 0.8 x 1032cm2 s-1 : mean nb. of int. = 2.3

                   prob. of 7 int. = 1%
           for L = 2.0 x 1032cm2 s-1 : mean nb. of int. = 5.4

                   prob. of 7 int. = 12%, of 8 8%
    Probability of N(interactions)/crossing

    For L = 0.8 x 1032cm2 s-1

     N(int)         36 bunch       99 bunch
      0           0.11219921     0.45137941
      1           0.24543345     0.35904841
      2           0.2684403      0.142802
      3           0.19573587     0.03786381
      4           0.10704204     0.00752966
      5           0.04683045     0.00119789
      6           0.01707344     0.00015881
      7           0.0053354      1.8046E-05
      8           0.00145888     1.7944E-06

    1 or more     0.88780079     0.54862059
    2 or more     0.64236734     0.18957218
    4 or more     0.17819117     0.00890638
    8 or more     0.00190983     1.9666E-06

      <N>         2.18747933     0.79544703
    Probability of N(interactions)/crossing
   For L = 2.0 x 1032cm2 s-1


     N(int)         36 bunch     99 bunch
      0           0.00421672    0.13688453
      1           0.02305996    0.27221098
      2           0.06305397    0.27066177
      3           0.11494105    0.17941425
      4           0.15714448    0.08919658
      5           0.17187515    0.03547558
      6           0.15665555    0.01175789
      7           0.122386      0.00334028
      8           0.08366151    0.00083032

    1 or more      0.99578328   0.86311547
    2 or more      0.97272333   0.5909045
    4 or more      0.79472831   0.14082848
    8 or more      0.18666714   0.00105815

       <N>        5.46869832    1.98861757
                      TFC timing
    From “minimum bias jet-brew” with L1 pt > 7GeV,
     get mean number and standard deviation for
     CTT tracks;
    parameterize Nt distribution by double
     exponential with  = 2 ;
    integration of this function allows estimate of
     P(Nt <16) for given number of overlapping
     interactions

    Nint      (Nt)         (Nt)      P(Nt < 16)
    1          0.8          1.5           1
    2          1.0          1.9           1
    4          1.1           2.1         0.997
    6          2.0           4.3         0.981
    8          3.7           6.0         0.957

    for L = 2.0 x 1032cm2 s-1 , and 396ns bunch
     crossing time, prob. of > 7 interactions is 18%; for
     8 interactions, prob. of more than 16 tracks =
     4%;
           negligible effect on deadtime.
               Simulation summary
   STT trigger simulation tools close to being ready
    and useful;
       tracking part operational;
       alternative clustering schemes to be implemented;
       LUT for hitfilter to be done
       work on providing test output in progress;
       caveat: need to ensure that algorithms in trig.
        simulation correspond to firmware implemented in
        hardware.


   queueing simulation package operational
       have STT queueing model which is quite realistic
       studies so far show no major bottleneck in design;
          even for high luminosities and occupancies;
       can easily adapt to new specifications as design
        progresses;
       more detailed TFC simulation in progress

								
To top