The ATLAS Trigger: HighThe ATLAS Trigger: High-Level

The ATLAS Trigger: High Level TriggerThe ATLAS Trigger: High-Level Trigger Commissioning and Operation During Early

Data TakingData Taking

Ricardo Gonçalo, Royal Holloway University of LondonRoyal Holloway University of London

On behalf of the ATLAS TDAQ High Level Trigger groupOn behalf of the ATLAS TDAQ High-Level Trigger group

EPS HEP2007 – Manchester, 18-25 July 2007

brought to you by C

OR

EV

iew m

etadata, citation and similar papers at core.ac.uk

provided by CE

RN

Docum

ent Server

https://core.ac.uk/display/44187218?utm_source=pdf&utm_medium=banner&utm_campaign=pdf-decoration-v1

Outline

The ATLAS High-Level Triggerg ggOverall system designSelection algorithms and steering

Trigger strategy for initial runningTrigger algorithm organisationTrigger strategy for initial runningTrigger strategy for initial runningStatus

High-Level Trigger Commissioning Technical runs Cosmic-ray runsy

Summary and outlook

Ricardo Goncalo, Royal Holloway University of London2ATLAS HLT Operation in Early Running

The ATLAS High-Level Trigger


Calo MuTrChOth d t t

Trigger DAQThree trigger levels:Level 1:

40 MHz

LVL1 2.5 μsCalorimeter Muon

MuTrChOther detectors

1 PB/sPipelines2.5 μs

Hardware basedCalorimeter and muons onlyLatency 2.5 μsOutput rate ~75 kHz

75 kHzLVL1 Acc.

ROD ROD ROD

CalorimeterTrigger

MuonTrigger

120 GB/sRoI’s

CTP

(Region of Interest)

Output rate ~75 kHz

Level 2: ~500 farm nodes(*)Only detector ”Regions of

ROB ROB ROB

120 GB/s

LVL2 ~40msL2SVROIB

RoI’s (Region of Interest)

RoI reques

ts

Only detector Regions of Interest” (RoI) processed -Seeded by level 1Fast reconstructionAverage execution time ~40 ms(*)

H

L2 kHz

RoI data

Event Builder3 GB/s

ROSL2PL2N

L2PL2P LVL2 Acc.

Average execution time 40 ms( )Output rate up to ~2 kHz

Event Builder: ~100 farm nodes(*)

T2 kHz

EBEvent Filter

EFPEFP

EFP

~4sec

EFNEF Acc.

( )

Event Filter (EF):~1600 farm nodes(*)Seeded by level 2

200 HzEFN

300 MB/sEvent Size ~1.5 MB

Potential full event accessOffline algorithmsAverage execution time ~4 s(*)Output rate up to ~200 Hz


300 MB/sOutput rate up to 200 Hz

(*) 8CPU (four-core dual-socket farm nodes at ~2GHz

Selection method EMROILevel1 Region of Interest is found and position in EM

l i i dL2 calorim.

cluster?

calorimeter is passed to Level 2Event rejection possible at each step

Electromagnetic

L2 trackingLevel 2 seeded by Level 1Fast reconstruction

Electromagneticclusters

match?

track?Fast reconstruction algorithms Reconstruction within RoI

E.F.calorim.

E F t kiE.F.tracking

track?Ev.Filter seeded by Level 2Offline reconstruction l ithalgorithms

Refined alignment and calibration

e/γ reconst.


e/γ OK?

Steering EMROI

Algorithm execution managed by Steering Based on static trigger configuration

L2 calorim.

cluster?

L2 calorim.

cluster?And dynamic event data (RoIs, thresholds)

Step-wise processing and early rejectionL2 tracking

p p g y jChains stopped as soon as a step failsReconstruction step done only if earlier step successful match?

track?

Event passes if at least one chain is successful E.F.calorim.

E F t ki

E.F.calorim.

Prescale (1 in N successful events allowed to pass) applied at end of each level

E.F.tracking

track?

Specialized algorithm classes for all situations

Topological: e g 2 μ with m ~ m

e/γ reconst.e/γ reconst.


Topological: e.g. 2 μ with mμμ ~ mZ

Multi-objects: e.g. 4-jet trigger, etc……e± OK?γ OK?

Trigger Strategy for Initial Running

Trigger algorithms

High-Level Trigger algorithms organised in groups (“slices”):Minimum bias, e/γ, τ, μ, jets, B physics, B tagging, ET

miss, cosmics, plus combined-slice algorithmsslice algorithms

For commissioningCosmics slice used to exercise trigger – already started!

For initial running:Crucial to have minimum bias, e/γ, τ, μ, jetsB physics will take advantage of initial low-lumi conditions (not bandwidth-critical)B physics will take advantage of initial low-lumi conditions (not bandwidth-critical)

Lower event rate allow low transverse momentum thresholds needed for B physics

ETmiss and B-jet tagging will require significant understanding of the detector

Will need to understand trigger efficiencies and rates using real dataZero bias triggers (passthrough)Minimum bias: 1 Select good offline Z→μμ/eeMinimum bias:

Coincidence in scintillators placed in front of calo.Counting inner-detector hits

Prescaled loose triggers“T d b ” th d t

1. Select good offline Z→μμ/ee 2. Randomly select “tag” lepton;

if triggered, use second lepton as “probe”


“Tag-and-probe” method, etc 3. ε = #(triggered probes)/#(all)

Trigger strategy for initial running

Major effort ongoing to design a complete trigger list (“menu”) for initial runningCommissioning of detector and trigger; early physicsCommissioning of detector and trigger; early physicsStart with L=1031 cm-2s-1 benchmark and scale accordingly

Many sources of uncertainty:Background rate (dijet cross section uncertainty up to factor ~2)Beam-related backgroundsNew detector: alignment, calibration, noise, Level 1 performance (calo isolation?), etcEvent occupancyEvent occupancy

Must be conservative and be prepared to face much higher rates than expected

Need many “handles” to understand the trigger:Many low-threshold, prescaled triggers, several High Level triggers will run in “pass-through” mode (take the event even if trigger rejects it)Monitoring framework (embedded in algorithms, flexible and with small overheads)Monitoring framework (embedded in algorithms, flexible and with small overheads)Redundant triggers

e.g. minimum bias selection with inner detector and with min.bias scintillators

E t th t l idl i ll it f l d t


Expect the menu to evolve rapidly, especially once it faces real data

Status

Trigger information routinely available in simulated dataTrigger decision and reconstructed objects easily accessible in simulated dataG t d h k d f db k f h iGenerated much work and feedback from physics groups

Trigger decision can be re-played with different thresholds on already reconstructed data: important for optimisation of selectionreconstructed data: important for optimisation of selection

Tools being developed for trigger optimisationEstimate efficiency, rate and overlapsy pNeed to be able to react quickly to changing luminosity conditions

A draft menu exists with some 90 triggersM h k i d i i i d i i h d di iMuch work is under way to optimise it and test it against the expected conditions

Rates, efficiencies and overlap between selections being studied for the menuIncluding misaligned detector in simulationIncluding misaligned detector in simulationIncluding overlapped events per bunch crossingIncluding natural cavern radiation (for muons)


High-Level Trigger Commissioning

Technical runs

A subset of the final High-Level Trigger CPU farm and DAQ system were exercised in “technical runs”

Simulated (Level 1 triggered) Monte Carlo events in raw data format preloaded into DAQ readout buffers and distributed to farm nodes

Realistic trigger list used (e/γ, jets, τ, B physics, ETmiss, cosmics)

HLT algorithms, steering, monitoring infrastructure, configuration d t bdatabase

Measure/exercise:

Data: 4

•Mean 31.5 ms/event•98% rejection @ L2•Average 40ms/ev available

Event latenciesAlgorithm execution timeMonitoring frameworkC fi ti d t b

40% W

/Z +

Configuration databaseNetwork configurationRun-control

+ 60% di-jmean 94.3 ms/event

Accepted events only


ets

Cosmics runs

A section of the detector was used in cosmics runs (see previous talk) including:including:

Muon spectrometer Tile (hadronic) calorimeterLAr (electromagnetic) calorimeterInner detector

The High-level was exercised successfully on real data in test cosmic runs


runs.

Conclusions and outlook

Conclusions and outlook

The ATLAS High-Level Trigger is getting ready to face LHC datag g y

The final High-Level Trigger system was successfully exercised in technicalwas successfully exercised in technical runs on simulated data and was shown to be stable

High-Level Trigger algorithms and machines took part in cosmics test runs

Trigger information now routinely available in simulated data

U d f t i ti i tiUsed for trigger optimisation

Looking forward to triggering on LHC


g gg gdata next year!

Backup slides


~ 500~1600

SDX1 dual-CPU nodesCERN computer 6 ~100

Trigger / DAQ architecture

SDX1

Second-leveltrigger

500

LVL2 farm

EventFilter(EF)

1600computer centre Event rate

~ 200 HzData storage

6Local

StorageSubFarmOutputs

(SFOs)

100 EventBuilderSubFarm

Inputs

(SFIs)

SDX1

et

LVL2S

Network switches

pROS

stores LVL2output

DataFlowManager

Network switches

USA15Gig

abit

Eth

erne

data

requ

ests

e co

mm

ands

ed e

vent

dat

a

Super-visor

E t d t USA15

Eve

nt

Del

ete

Req

uest

e

ns O

f Int

eres

t

Data of events acceptedb fi t l l t iUSA15 1600

Event data pulled:partial events @ ≤ 100 kHz,

Reg

ion

UX15

Read-Out

Dedicated linksVME

by first-level trigger

Read-OutSubsystems

USA15 1600Read-OutLinks

~150PCs

@ ,full events @ ~ 3 kHz

Drivers(RODs) First-

leveltrigger

Subsystems(ROSs)

RoIBuilder

Timing Trigger Control (TTC)


Event data pushed @ ≤ 100 kHz, 1600 fragments of ~ 1 kByte each

Configuration

Trigger configuration:Active triggersTh i tTheir parametersPrescale factorsPassthrough fractionsConsistent over three trigger levelsgg

Needed for: Online running Event simulationEvent simulationOffline analysis

Relational Database (TriggerDB) for online runningonline running

User interface (TriggerTool) Browse trigger list (menu) through keyRead and write menu into XML formatMenu consistency checks

After run, configuration becomes conditions data (Conditions Database)


conditions data (Conditions Database)For use in simulation & analysis

Single-e Tr. Eff. (from Z→e+e-)f ti f φ d Eas a function of η, φ and ET

Mi li d G tMisaligned Geometry

Wrt. offline:Loose electron


Tight electron

Trigger efficiency from data

Electron trigger efficiency from real Z→e+e- data:

MZ (GeV)

real Z→e e data:1. Tag Z events with single

electron trigger (e.g. e25i) 2. Count events with a second

electron (2e25i) and mee ≅ mZ

N d d f dNo dependence found on background level (5%, 20%, 50% tried)

~3% statistical uncertainty after 30 mins at initial luminosity

Small estimated systematic uncertainty Method Z→e+e- counting

Level 2 efficiency 87 0 % 87 0 %


Level 2 efficiency 87.0 % 87.0 %

Trigger pT threshold(*) Obs

Σ ET (jets) ? ?

Trigger pT threshold(*) Obs

Electron 5,10,15, Prescale

ETmiss 12, 20, 24, 32,

36, 44Prescale

ETmiss 52, 72 No presc

J/Ψ→ee Topological B-phys

Electron 20,25,100 No presc

Di-electron 5,10 Prescale

Di-electron 15 No prescJ/Ψ→ee Topological B phys

μ μ 4 B-phys

J/Ψ→ μ μ Topological B-phys

BsDsPhiPi Topological B-phys

Photon 10,15,20 Prescale

Photon 20 No presc

Di-photon 10 Prescale

Di photon 20 No presc p g p y

BγX B-phys

e + ETmiss 18+12 Prescale

μ + ETmiss 15+12 No presc

Di-photon 20 No presc

Jets 5,10,18,23,35,42,70 Prescale

Jets 100 No presc

3 Jets 10 18 B-tagJet + ET

miss 20+30 No presc

2 Jets + ETmiss 42+30 No presc

Jet+ ETmiss +e 42+32+15 No presc

3 Jets 10,18 B-tag

4 Jets 10, 18 B-tag

4 Jets 23 Express

τ 10, 15, 20, 35Jet+ ET

miss + μ 42+32+15 No presc

4 Jet + e 23+15 No presc

4 Jet + μ 23+15 No presc

τ 10, 15, 20, 35

Di- τ 10+15,10+20,10+25

Muon 4, 6, 10, 11, 15, 20, 40 Muon spectr

τ + ETmiss 15+32,25+32,

35+20,35+32

τ + e 10+10 Express

τ + μ 10+6 Express

spectr.

Muon 4, 6, 10, 11, 15, 20, 40 ID+Muon

Di-muon 4, 6, 10, 15, 20 Passtthr.

ΣET 100 200 304 prescale


τ + μ 10+6 Express

2 τ + e 10+10 Express

ΣET 100, 200, 304 prescale

ΣET 380 No presc


Documents

The ATLAS Trigger: HighThe ATLAS Trigger: High-Level