New mixing in Fast Simulation

Introduction

A new mixing scheme (at RECO-level for tracks, at SIM-level for calorimeter and muon hits) is implemented, but not available by default yet.

Mixing happens at two levels, with SIM and RECO input: in the digitizers of ECAL, HCAL, Muon detectors (both in-time and out-of-time, eventually) and after tracking (only in-time).

The idea is then, once everything is fully validated, to split the jobs into two steps, with the goal of mitigating the peak memory usage.

Instructions to test the new mixing

cmsrel CMSSW_7_1_0_pre9
cd CMSSW_7_1_0_pre9/src
git cms-addpkg FastSimulation/Configuration
git cms-merge-topic 4179
scram b -j 8

edit FastSimulation/Configuration/python/CommonInputs_cff.py and set MixingMode = 'DigiRecoMixing' instead of 'GenMixing'

cp /afs/cern.ch/user/g/giamman/public/test.py .  # this only saves the track collections

# this cmsDriver command has the only PU scenario adapted so far:
cmsDriver.py SingleMuPt10_cfi.py -s GEN,SIM,RECO --fast --pileup=mix_2012_Summer_inTimeOnly --conditions auto:startup_GRun --beamspot Realistic8TeVCollision --eventcontent=FEVTDEBUGHLT --datatier GEN-SIM-DIGI-RECO -n 10

A private FullSim minbias with RECOSIM event content: /afs/cern.ch/user/g/giamman/public/mixing/MinBias_GENSIMRECO.root

Status

Calorimeters and muons: everything seems ok so far, including OOT PU

Tracks: mixing happens. Tracks get propagated to the rest of RECO.

Performance shown by Gustavo to be better than the old mixing in terms of cpu and memory (even without job splitting).

Private validation performed by Viktoria with PFjets.

Latest update in this talk

How to create a minbias sample for mixing (on top of 7_1_0_pre9)

New event content entered here:

git cms-merge-topic 4197
Create MinBias sample with:
cmsDriver.py MinBias_13TeV_cfi -s GEN,SIM,RECO --conditions auto:startup_GRun --eventcontent=FASTPU -n 100 --fast --fileout MinBias_for_mixing.root

How to create a minbias sample for mixing (obsolete, kept for the size estimation)

Minimum requirement is that the rootfile must contain the SimHits, the Tracks, the TrackExtras, the Trajectories, and the TrackingRecHits.

This command provides a file with sufficient information to be then used as input for pileup mixing (+):

cmsDriver.py MinBias_13TeV_cfi -s GEN,SIM,RECO --conditions auto:startup_GRun --eventcontent=FEVTDEBUG -n 100 --fast --fileout MinBias_for_mixing.root

The size of its output is roughly 330 kB per event. How many minbias events are used for FullSim relvals?

In alternative, one can add this block to FastSimulation/Configuration/python/EventContent_cff.py:

#####################################################################
#
# To be used only to create the MinBias sample for "new mixing" (--eventcontent=FASTPU)
#
#####################################################################
FASTPUEventContent = cms.PSet(
    outputCommands = cms.untracked.vstring('drop *', 
        'keep *_famosSimHits_*_*',
        'keep *_MuonSimHits_*_*',
        'keep Traj*_*_*_*',
        'keep *_generalTracks_*_*')
)

and then run:

cmsDriver.py MinBias_13TeV_cfi -s GEN,SIM,RECO --conditions auto:startup_GRun --eventcontent=FASTPU -n 100 --fast --fileout MinBias_for_mixing.root

The size of the output is now roughly 106 kB per event (77 kB without Traj*_*_*_*).

If one also drops the SimHits in the tracker (*_famosSimHits_TrackerHits_*), the size drops to 82 kB per event (including Traj*_*_*_*).

(+) edit FastSimulation/Configuration/python/mixHitsAndTracksWithPU_cfi.py to change input pileup file.

To do

- Integrate with the relval production (It implies a central production for proper minbias)

- Mix FastSim with FullSim

- Check that reco can be run on a step of its own, i.e., that all of its inputs are saved by the digitization sequence

- Job splitting: test simu+digi followed by reco (scenario 1) versus simu followed by digi+reco (scenario 2); ask Gustavo to do performance report, and Victoria to check with a decent statistics that the physics is the same

Notes about job splitting

- cmsDriver must not add Playback = True to the step2 configuration

- is it more convenient to have DIGI (with mixing module) in the first or in the second step?

Topic attachments
I Attachment History Action Size Date Who Comment
GIFgif dt_digi_0pu.gif r1 manage 4.3 K 2013-03-03 - 15:49 AndreaGiammanco  
GIFgif dt_digi_10pu.gif r1 manage 4.0 K 2013-03-03 - 15:49 AndreaGiammanco  
GIFgif dt_digi_summer12.gif r1 manage 3.7 K 2013-03-03 - 15:49 AndreaGiammanco  
GIFgif hcal_digi_0pu.gif r1 manage 5.6 K 2013-02-28 - 17:24 AndreaGiammanco  
GIFgif hcal_digi_10pu.gif r1 manage 6.6 K 2013-02-28 - 17:24 AndreaGiammanco  
GIFgif hcal_digi_summer12.gif r1 manage 5.8 K 2013-02-28 - 17:24 AndreaGiammanco  
GIFgif ntk_0pu.gif r1 manage 4.2 K 2013-04-12 - 11:00 AndreaGiammanco  
GIFgif ntk_10pu.gif r1 manage 6.4 K 2013-04-12 - 11:00 AndreaGiammanco  
GIFgif ntk_2012pu.gif r1 manage 6.3 K 2013-04-12 - 11:00 AndreaGiammanco  
GIFgif nvtx_0pu.gif r1 manage 4.3 K 2013-04-12 - 11:00 AndreaGiammanco  
GIFgif nvtx_10pu.gif r1 manage 5.8 K 2013-04-12 - 11:00 AndreaGiammanco  
GIFgif nvtx_2012pu.gif r1 manage 6.2 K 2013-04-12 - 11:00 AndreaGiammanco  
GIFgif sumCaloET_0pu.gif r1 manage 5.9 K 2013-03-03 - 11:42 AndreaGiammanco  
GIFgif sumCaloET_10pu.gif r1 manage 6.3 K 2013-03-03 - 11:42 AndreaGiammanco  
GIFgif sumCaloET_Summer12.gif r1 manage 5.3 K 2013-03-03 - 11:42 AndreaGiammanco  
GIFgif trackEffNoCF.gif r1 manage 4.0 K 2013-03-13 - 14:22 AndreaGiammanco  
GIFgif trackEffWithCF.gif r1 manage 3.9 K 2013-03-13 - 14:22 AndreaGiammanco  
Edit | Attach | Watch | Print version | History: r57 < r56 < r55 < r54 < r53 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r57 - 2014-06-16 - AndreaGiammanco
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    CMSPublic All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback