Software and Simulation Status

CBM Collaboration Meeting, GSI, 13 March 2009

Volker Friese

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese2

Status detector simulations

Transport

CbmMCPoint

HitProducer

CbmHit

Track Finder

Transport

CbmMCPoint

Digitiser

CbmDigi

Track Finder

HitFinder

CbmHit

Position smearing,

independent points

Detector response model

Interaction of points

simulation

reconstruction

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese3

Status detector simulations: MVD

No Electric Field:

Electrons are diffusing

θ

sensitive volume

C. Dritsa

Model for charge production and diffusion in sensor2D cluster of fired pixelsCross-check with prototype measurementsCluster finder

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese4

Status detector simulations: STS

Model for charge production and diffusion in sensor1D cluster of fired stripsCluster finding (centre of gravity)Hit finding (crossing of front/back cluster centres)

A. Kotynia

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese5

Status detector simulations: ECAL

Very detailed studies on photon reconstruction

Shower library developed

Problem of merging clusters

M. Prokudin

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese6

Status detector simulations

• With MVD and TRD, detector response models are now implemented for all subsystems

• Tuning to prototype data to come once available

• Study and tuning of detailed detector and FEE properties now accessible

• Latest developments (MVD, STS, TRD) not yet taken into account in tracking / physics simulations

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese7

Observables: open charm

Extensive studies by I. Vassiliev in many channels

Signals observed over background in all cases

Requires 1st MVD @ 5 cm

To be checked:

influence of clustering in MVD, STS

delta electrons

pile up (up to 10 tolerable?)

I. Vassiliev

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese8

Observables: neutral particles in ECAL

Reconstruction of vertex γ, π0, η studied Reasonable reconstruction efficiency obtained π0 signal clearly visible above background η requires more simulation statistics (O(106)

events) Low-mass background studied (A. Stavinskiy)

S. Kiselev

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese9

Observables: Flow (event plane angle resolution)

Event plane reconstruction (and centrality selection) using PSD information

First results promising (resolution 40o – 50o) Needs to be checked for non-zero event plane

angles Possibly requires selection of neutrons in PSD Similar studies by S. Seddiki, A. Maevskaya

V. Pozdniakov

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese10

Algorithms and methods: wavelets

Promising method for detection of noisy signals

New application examples shown Possible application: Extracting signal yields

for small S/B ratios (no BG subtraction needed)

G. Ososkov

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese11

Algorithms and methods: particle ID in TRD and TOF

• Application of statistical criteria (likelihood, mena value, ωkn) in TRD was studied

and compared to ANN.

• Performance of ANN was found superior.

• However, cross-check with independent method is desirable.

• Online implementation to be investigated (J/ψ trigger)

• Difference of dE/dx GEANT/prototype deteriorate the electron ID performance.

• First application of ωkn method to TOF hadron ID (requires two independent TOF

measurement). Preliminary results require further investigations.

O. Denisova

T. Akishina

V. Ivanov

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese12

Trigger studies: open charm

L1CATrackFinder

L1KFTrackFitter

Charm Track Candidates Selection χ2prim > 3

Charm Pairs χ22geo

< 3.0, zv <1 cm χ2

topo < 3.0, minv > 1.3 GeV

Charm Triplets χ23geo+topo < 3.0 ☺D+c Ds

☺D0☻☻

I. Vassiliev

Trigger algorithm developed Requires 1st MVD @ 5cm Rejection factors O(100) achievable

w/o loss of signal (w.r.t. offline analysis)

Requires (full) STS reconstruction, but only reduced combinatorics due to selection on single-track level

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese13

Trigger studies: charmonium (e+e-)

Simple trigger logic: Require >2 tracks/event with pt > 1 GeV, identified as electrons in TRD

A. Maevskaya

MC PIDMC PID Like>0.4Like>0.4 Ann>0.5Ann>0.5

Target 250Target 250μμ 1428514285 559559 119119

Target 25Target 25μμ 2000020000 813813 172172

No loss of signal w.r.t. offline analysis Requires:

(full) STS reconstruction partial TRD reconstruction electron ID in TRD (ANN / statistical)

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese14

Trigger studies: charmonium (μ+μ-)

x=0,y=0x=0,y=0

∆x,∆y∆x,∆y Trigger strategy:

Have two tracks after last absorber Fit triplet and extrapolate back to target Cut on distance to target

Requires: information only from last three detector stations

Can be improved by using TOF (2nd level?)

segmentation

trigger without χ2 xz=0 yz=0

εJ/ψmBias, % 23.4 20.3 15.5 15.2

background

suppression factor (bsf)

for mbias events

1 1 318 606

A. Kiseleva

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese15

Reconstruction: Hough Transform

Z

X

Y

P y/P z

P x/P z

1 /P z

Algorithm is being implemented on CellBE (Sony Playstation III) as prototyping system of FPGA array

Hough transform is calculated offline through LUT: details of field and geometry are uncritical

C. Steinle

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese16

Reconstruction: L1

Real-time performance on the quad-core Xeon 5345 Real-time performance on the quad-core Xeon 5345 (Clovertown) at 2.4 GHz – speed-up 30 with 16 threads(Clovertown) at 2.4 GHz – speed-up 30 with 16 threads

CPU/GPUCPU/GPU AMD: AMD: FusionFusion

CPU/GPUCPU/GPU AMD: AMD: FusionFusion

OpenCL?OpenCL?OpenCL?OpenCL?

GamingGaming STI: STI: CellCell

GamingGaming STI: STI: CellCell

GP CPUGP CPU Intel: Intel: LarrabeeLarrabee

GP CPUGP CPU Intel: Intel: LarrabeeLarrabee

GP GPUGP GPU Nvidia: Nvidia: TeslaTesla

GP GPUGP GPU Nvidia: Nvidia: TeslaTesla

CPUCPU Intel: Intel: XXX-coresXXX-cores

CPUCPU Intel: Intel: XXX-coresXXX-cores

FPGAFPGA XilinxXilinx

FPGAFPGA XilinxXilinx

I. Kisel

Impressive gain by vectorisation and multi-core architecturs

Future paths not clear, but computing paradigm will surely be parallelisation

Impact on our computing / software model? Migration from C++ to ???

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese17

Analysis: computing model

• It is yet unclear whether the physics analysis of CBM data will be done

– on a world-wide grid (LHC-like)

– on a small number of supercomputing centres (Frankfurt, ...)

• For the medium-term simulations, we will use CBM-GRID where necessary

• Set-up done by F. Uhlig (currently GSI only); first SIM+RECO run done successfully (Jan. 2009)

• Next step: JINR-LIT; ressources deployed

CBM-GRID user tutorialthis afternoon

You are welcome!

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese18

Reconstruction: computing model

• Raw date size: 5 PB / CBM run year

• Conventional approach: Several reconstruction runs with improved detector understanding / alignment / calibration befor physics analysis

• 2009 core time per min. bias Au+Au event: ≈ 10 s

• Core time per reconstruction run: 6 · 106 d

• Number of cores required (target: 100d / run): 6 · 104

• Will most probably executed on the same farm as online event selection

• Can, be proper means, the complete reconstruction be made fast enough to be performed on-line?

CBM request for POF II:2 · 104 core days (2009)+ 100% annual growth20 TB storage (2009) + 100% annual growthTarget (2014): 10% of full ressources

CBM Collaboration Meeting, Darmstadt, 13 March 2009 Volker Friese19

Next steps?

• Physics performance:– Go beyond simple S/B as quality check

– Simulate spectra to assess the performance for a given physics goal

– Determine number of needed events for this goal

– Arrive at a runtime scenario for each physics observable

– CBM Physics Performance Report: 2012/2013

• Short-term priority: Simulations for CBM@SIS100– A+A, 2-10 AGeV

– p+p, p+A, 2-30 GeV

• Demonstrate feasibility of event reconstruction from free-streaming raw data