Simulation overview and resource needs Jin Huang BNL June 18 2018 BNL June 18 2018 sPHENIX Software amp Computing Review 1 Overview for this talk June 18 2018 sPHENIX Software amp Computing Review ID: 798633
Download The PPT/PDF document "sPHENIX Software & Computing Review" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
sPHENIX Software & Computing Review
Simulation overview and resource needsJin Huang (BNL)June 18, 2018BNL
June 18, 2018
sPHENIX Software & Computing Review
1
Slide2Overview for this talkJune 18, 2018
sPHENIX Software & Computing Review2
EMCalInner HCalMagnetOuter HCal
IP
Geant4 Land
In this talk:
Simulation resource consumption
Expected major simulation samples
Question
:
Are the resources required to process the data to a form suitable for physics analyses adequately understood?
Detailed setup in Geant4 integrated with full event
reco
Test beam verifications
Frequently used in analysis by the collaboration for 1M Au+Au full event + embedding simulations.
Slide3Simulation framework integrationGeant4 simulation and reconstruction are integrated in the sPHENIX software framework
More on framework see last talk by Chris P.In production mode: run Geant4 sim in central production (CPU intensive), buffer the output file (DST) for reuse (require disk space), then run reconstruction in separated user sessions.Ensure same configuration and geometry are used in simulation and reconstruction → embedding For example, Geometry and magnetic field configuration in Geant4 is automatically passed down to reconstruction stage for use in alignment adjustment, in tracking Kalman
filter and in calorimetry geometric presentation. 3June 18, 2018Output Managers
Input Managers
Fun4AllServerdata objectsGeant4 sim Modules
User
Reconstruction Modules
Analysis Modules
Order of execution
Slide4Simulation chain4
100
ms / ev (HIJING)2000 s / ev100 ms / evSee reco. talks1.7 MB / ev * 29%210 MB / ev * 49%Not saved
See reco. talks
Above table : per-event resource for central Au+Au event in full sPHENIX. Store object optimized for fast dev, rather for space saving yet. Stored in ROOT file with compression ratio shown in the table.Expect order of 10M Au+Au event sim total (dominated by rare fake jet background simulation need, next slide...)Rare signal probes simulated in the embedding mode (fast 10s/ev, regularly carried out 0.1-1M sample studies in the past)
Slide5Required data samplesLeading requirement on simulation computing need driven by simulating Au+Au events in search for rare background. In particular fake jet background in central Au+Au collisions
Dominating simulation resource needed: 10M central Au+AuFake jet is important up to ET = 30-35 GeV, 10% fake jet stat. precision in ET = 30-35 GeV, require 100 fake jets 100 fake jets @ 30-35 GeV require 100 fake jet/5GeV/(10-5 jet/GeV/
ev) = 2 M central Au+Au eventsOver design and early analysis of sPHENIX, expect run this study over five iterations: 5x2 M central Au+Au simulationsSame sample will be used in rare signal embedding (fast), (Upsilons→e+e-) background embedding+mixing, and test reconstruction software (see earlier talks)5x 2M central Au+Au translate to 600 CPU-Core*year and 200 TB storage for each of the 5 iterationsSmall compare to sPHENIX reconstruction need (last three talks)5June 18, 2018sPHENIX Software & Computing Review
DOI: 10.1103/PhysRevC.86.024908
1x 34-35 GeV fake jetper 100k eventConsistent with the order to observe5-sigma fluctuation in all R=0.4 cones
Slide6Simulation Setup: Tracker
6
sPHENIX Software & Computing ReviewDesign to SimulationGeometric parameters, materials, support GDML import
MVTX Ladders modeled in details
CAD
Model
sPHENIX Geant4 display of
p
T
=30 GeV/c
B
+
-hadron
GenFit
track/
sens.
Vol. display
Automatic geometry and field map porting
to reconstruction
Used in
Kalman
filter
Reco
event display
Consistent way to apply alignment
and embedding in future
Inner tracker (MVTX and INTT)
Slide7Simulation Setup: EMCalJune 18, 2018
sPHENIX Software & Computing Review7
10GeV, e+2 cm SPACAL Towerw/ fibers displayedTotal 20 M fiber simulated each event
EMCal Half Sector with possible
η-extension(fibers simulated but hidden from display)full EMCal
Slide8Simulation Setup: HCalJune 18, 2018
8
Baseline design24-GeV pion showerEMCal
Inner HCal
MagnetOuter HCalBeam view of full calorimetersCalorimeter simulation & analysis chain: GEANT4 hit → Scintillation light model → Light collection model → Tower readout → Digitization → Calibrated tower energy → Clustering/Track matching/Jet findingsPHENIX Software & Computing Review
Slide9Test Beam Verification [arXiv:1704.01461]June 18, 2018
sPHENIX Software & Computing Review9
Geant4 World
Slide10Test Beam Verification [arXiv:1704.01461]June 18, 2018
sPHENIX Software & Computing Review10
EMCal energy resolution for EM shower in tower centerHCal energy response for pion-New 2017 data2016 data and submitted to IEEE TNS
Slide11Au+Au background → Compact showerJune 18, 2018
sPHENIX Software & Computing Review11
EMCalInner HCal(if instrumented)Outer HCalsPHENIX
simulation
sPHENIX simulationsPHENIX simulation
Slide12Example simulation campaign : Jet Observables
Jet-balance/imbalance observables simulated in full detector events: day-1 measurement, resolution under controlOn-going effort in understanding the details of the jet simulations and unfolding studies.
Expanding jet observables studied with the updated design and simulationsJune 18, 2018sPHENIX Software & Computing Review12Di-jet transverse momentum asymmetryγ-jet transverse momentum balanceShape of observable well controlled from generated shape → measured quantitiesPromising for unfolding and sys. control
γ-jet fragmentation function
Slide13Example simulation campaign : UpsilonJune 18, 2018
sPHENIX Software & Computing Review13
Upsilon simulation involves simulation and reconstruction of tracking, calorimetry, embedded signal and backgroundsUpsilon→ , hadron/electron ID simulated via embedding in full event-full detector Geant4 simulations. Possible for real AuAu data embedding after sPHENIX turn-onFast simulation with G4-extracted e-ID to produce full spectrum line shape for
Pion rejection @ 90% efficiency(higher than proposal assumption)Upsilon spectrum in Au+Au
(after combinatorial subtraction)
Test beam: EMCal
rejection for pion-
Slide14SummarysPHENIX has mature and detailed simulation, verified in test beam, regularly used with Million-embedded-event simulation campaigns by the collaborationFor simulating one central Au+Au event: 2000s CPU-core + 200MB disk
CPU time dominated by shower simulation in detailed calorimetric structures to represent rare tails, fluctuation and non-compensating hadron calorimetryCPU-save steps by reusing Geant4 result files and embedding of signalsMay further optimize with fast simulation for specific studiesDisk storage object not yet optimized for space saving. Dominated by Geant4 hit info and truth tracing tablesDominating computing need for simulation: 5 iterations of 2 M central Au+Au simulation to study rare backgroundRequire 600 CPU-Core*year and 200 TB storage for each of the 5 iterations
Small compared with computing need for sPHENIX reconstruction14June 18, 2018sPHENIX Software & Computing ReviewQuestion: Are the resources required to process the data to a form suitable for physics analyses adequately understood?
Slide1515June 18, 2018
sPHENIX Software & Computing Review
Back Up
Slide16June 18, 201816
EMCal radius (cm)
EMCal X (cm)EMCal Y (cm)← Moliere radius, 2cm
← 3x3 tower
95% containmentEMCal radius (cm)Inner HCal radius (cm)EMCal X (cm)Inner HCal X (cm)Inner Hcal Y (cm)EMCal Y (cm)
← 3x3 tower
50% containment
← ~3x3 tower
60% containment
― Energy deposition (A.U.)
― Percentage outside radius
4 GeV Electrons in EMCal
4 GeV
Pions
in EMCal
, that
passed E/p electron-ID cut
EMCal
EMCal
Inner HCal
(if instrumented)
A.U.
/
Leakage Ratio
A.U.
/
Leakage Ratio
A.U.
/
Leakage Ratio
sPHENIX Software & Computing Review
Slide17Performance : Single EM Showers
June 18, 2018sPHENIX Software & Computing Review17
Beam test [arXiv:1704.01461] Full sPHENIX detector simulationApplied cluster-position-based non-uniformity correction as used in test beam analysisdE/E < 16%/ √(E) + 5%
Slide18Performance : Photon in Full Event
Good linearity up to photon kinematic limit of sPHENIXSatisfying UPP-
: ΔE/E <8% for E>15 GeV in central Au+Au collisions June 18, 2018sPHENIX Software & Computing Review18
ΔE/E <8% @ E = 15 GeV
Slide19Performance : Jet FindingJets in p+p and central Au+Au collisions are also studied in full detector simulationsJet finding followed by ATLAS style iterative background subtraction [
10.1103/PhysRevC.86.024908]Performance meeting sPHENIX UPP-Jet: ΔET
/ET <150%/√ET in central Au+Au collisionsFurther improving underlying event subtraction and fake-jet rejection algorithm based on RHIC and LHC experiencesJune 18, 2018sPHENIX Software & Computing Review19sPHENIX UPP-Jet