Post on 01-Sep-2018
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 1
High-Level Trigger Studies
Darin AcostaUniversity of Florida
DOE/NSF Report on US CMSSoftware and Computing
DOE GermantownJune 29, 2000
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 2
CMS DAQ ArchitectureCMS DAQ ArchitectureCMS has a multi-tiered trigger system:
è L1 reduces rate from 40 MHz to 75 kHzp Custom hardware processes calorimeter and muon data to
select electrons, photons, jets, muons, ET above threshold
è L2, L3,… (HLT) reduces rate from 75 kHz to 100 Hzp Commercial CPU farm runs online programs to select
physics channels
HLT
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 3
DAQ / HLT ChallengeDAQ / HLT Challenge
è HLT selection algorithms must keep only 1 eventout of 1000
p Limited by ability to filter a L1 data sample which is already rich in physics
è DAQ bandwidth is finitep Total event size is 1 MB, event rate is 75 kHzp Bandwidth limited by switch and link technology
First HLT milestone:
Demonstrate that L2 can achieve 10X rate reductionusing only 25% of the event data
è Partial event reconstruction using only calorimeterand muon data
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 4
CMS Physics GroupsCMS Physics GroupsPhysics Reconstruction and Selection (PRS)groups were established by CMS in 1999:
è Electron/Photon C. Seezè Muon U. Gaspariniè Jet/Missing ET S. Enoè b/tau A. Caner
p Overall coordination by P. Sphicas
Charge is to evaluate the physics capability of CMS from L1 to offline
US-CMS has substantial involvement in the Electron/Photon, Jet/MET and Muon groups
è Focus of the Calorimeter, Endcap Muon, andTriDAS construction project communities
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 5
Context of StudiesContext of StudiesThe L1 Trigger TDR is targeted for November, 2000
è Final design efficiencies and rates beingdetermined
The DAQ TDR is targeted for November, 2001è Need to understand the rejection capability of the
HLT triggers and the amount of data each steprequires to validate possible hardware solutions
A Physics TDR is targeted for 2003è It was delayed to allow for the transition to
object-oriented software
In all cases, we need to validate the algorithms for the CMS physics plan, taking into account all possible backgrounds
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 6
HLT HistoryHLT HistoryFirst HLT milestone was set for November 1999,6 months after the creation of the PRS groups
Despite rapid progress, milestone was not met:è Growing pains:
p First wide scale use of ORCA as a toolè Learning curve:
p Many inexperienced users and developers tackling C++è Not enough time:
p More validation needed of reconstruction softwarep Endcap Muon software not completedp HLT algorithms developed only at the last minute
è Insufficient manpower and resources
HLT milestone postponed until July 21, 2000
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 7
Recent DevelopmentsRecent DevelopmentsTransition from ORCA3 to ORCA4
è Full migration to Objectivity in place of Zebra filesè Additional functionality
p Inclusion of endcap muon code (L1 and reconstruction)p Correct handling of out-of-time b.x. in calorimeter
è Further validation testsProduction of large simulated event samples
è Why? σσinel = 55 mb and L = 1034 (⇒⇒ 17.3 collisions/b.x.)
è GEANT3 production at CERN, Italy, and U.S.(Caltech, UFlorida, UCDavis)
è OO hit formatting and digitization done at CERNp 2 weeks for each on a large Linux PC farm
è Users access a User Federationp All file/database details hidden from userp Standard ntuples produced for physics groups
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 8
ORCA Production 2000ORCA Production 2000Signal
Zebra fileswith HITS
ORCADigitization
(merge signal and MB)
ObjectivityDatabase
HEPEVTntuples
CMSIM
HLT AlgorithmsNew
ReconstructedObjects
MC Prod.
ORCA
Prod.
HLT
Grp Databases
ORCAooHit
FormatterObjectivityDatabase
MB
ObjectivityDatabase
Catalog import
Catalog import
ObjectivityDatabaseObjectivityDatabaseytivitcejbOytivitcejbOesabataDesabataD
Mirrored D
b’s(U
S, Russia, Italy..)
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 9
Data SetsData Setse/γγ and Jet/MET groups:
è 0.9 Mevt (0.8 TB) è Supplement 1999 Monte Carlo production
p 0.6 Mevt QCD dijet (various ET ranges)p H→→ bb, H→→ ττ ττ , SUSY, min. bias pile-up, single particle
Muon group:è 1.3 Mevt (1.3 TB) è Production redone in 2000 to update Pythia,
endcap geometry, and event weightingp 0.9 Mevt minimum bias events for rate and pile-upp H→→ ZZ →→ 4µµ , H→→ WW →→ 2µµ , W→→ µνµν , Z →→ µµµµ , single µµ
CMS operations:è Expect 1000 TB/year
HLT milestone is a 0.1% Mock Data Challenge!
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 10
Muon Physics PlanMuon Physics PlanValidate L1 trigger decision and use as a seed for muon reconstruction
Improve PT resolutionè Refit with vertex constraint
using reconstructed hitsè Extrapolate to outer tracker
and refit
Apply isolation requirementè Extrapolate to calorimeter
and sum energy in cone
U.S. physicists working on L1 and Reconstruction for EMU
Mu
on
Rat
eEffective PT Threshold
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 11
Results: Single Results: Single µµµµ L1 EfficiencyL1 Efficiency
First look with ORCA4
Recoverable
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 12
Barrel/Endcap Overlap RegionBarrel/Endcap Overlap Region
Cathode Strip Chambers
Drift Tubes
Must include ME2-ME3 tracks for redundancy, despite poor PTresolution
We are learning from the simulation…
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 13
e / e / γγγγ and Jet / MET Physics Planand Jet / MET Physics PlanElectrons:
è Bremsstrahlung recovery (>½X0 of silicon tracker)è ππ0 rejectionè Isolationè Track match and E/p cuts (L3)
Jets:è Finer tower granularityè Improved jet axisè Energy calibration (non-linearity, pile-up)
p All this improves the ET resolution MET:
è Improve the ET resolution Taus:
è Improve shape cuts
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 14
Jet Jet Resol’nResol’n before & after Correctionbefore & after Correction
Average for |ηη| < 3
Cone R=0.5 No Pileup Cone R=0.5 Pileup
June 29, 2000 DOE/NSF USCMS Computing and Software Report. HLT Studies D. Acosta 15
ConclusionsConclusionsClosing in on HLT milestone for L2 performance
è Expect results in mid – Julyè However, this is just a first pass at the code that will eventually run
online. Will have to optimize, etc.
The task is challenging and work intensive!è Production is finally complete, analysis is underway
The HLT groups are driving (US)CMS software & computing
Significant work performed by U.S. physicists in Endcap Muon, Jet/MET, and e/γγ groups
è Could use even more manpower, since the “users” are often the code developers as well
User facilities and software/computing support are essential for U.S. physicists to maintain the prominent role taken so far