Moore/MuId in HLT

Download Report

Transcript Moore/MuId in HLT

integration in HLT-status and prospects
M. Biglietti Universita’ di Napoli-Federico II
G. Cataldi - INFN Lecce
and the Moore group
17/07/2015
Gabriella Cataldi INFN Lecce
EF Algorithms



Operate in a way close to that of the offline
environment BUT
are called and steered by the Step Controller of the steering
sofware that replaces the Athena Event Loop manager
They should not operate in a general-purpose sense, but they
must be seeded

Validate or reject Trigger Elements (TE) formed at the previuos
stage
Region selector mechanism

Trasformation of input TEs into a new output TEs

collection of required TEs to be validated



use the trigger sequences
use the configuration signatures
may be executed N times for each event
 must have a latency of 1 s
17/07/2015
Gabriella
Cataldi INFN
Lecce
 possibility of full access to
event
data

Requests to Moore


Should be driven from the Step Controller in
paths of sequences
Should consider only relevant region-of-interest




Conversion of geometrical region (h, f) into hash
identifiers by means of a Region Selector
hash identifiers are related to a DetectorElement (RPC
and MDT)/offline identifiers
Requests from the MdtDigitContainer and
RpcDigitContainer only the corresponding collections
Cache mechanism for multiple execution on one
event
17/07/2015
Gabriella Cataldi INFN Lecce
Moore flow unseeded
MooAlgs
RPC/TGC digits 
MooMakePhiSegments
PhiSegments
MooMakeRZSegments

MDT digits
MooMakeRoads
CrudeRZSegments
MooRoads
MooMakeiPatTracks
MooStatistics
MooiPatTracks
MooMakeNtuples
Ntuples
17/07/2015
Gabriella Cataldi INFN Lecce

Each step is
driven by an
Athena topalgorithm
Transient objects
are passed via
TDS/StoreGate
Independent
algorithms, the
only coupling is
through the
transient objects
Strategy
for Moore Seeding
RecMuonRoI
(h ±Dh, f±Df)
TDS
(ZEBRA)
RPCDigitContainer
MDTDigitContainer
RegionSelector
Hash offline IDs
DigitsCollection
(RPCs and MDTs)
MooMakePhiSegmentSeeded
MooMakeRZSegmentSeeded
PhiSegmentContainer
RZSegmentContainer
(TDS)
17/07/2015
MooHLTAlgo
decision
Tracks
Gabriella Cataldi INFN Lecce
MooMakeRoads
MooMakeTracks
MuonIdentification
Athena Implementation
 Inputs from Moore, Calo Reco and iPat

Athena modules:

MuidInit :

gets tracks from Muon Reconstruction (Moore) and associates the
truth from KINE bank

MuidStandAlone:

muon tracks are propagated to the vertex
 multiple scattering parameterised as scattering planes in
calorimeters
 energy loss from truth and/or from Calo Reconstruction (Tile,
HECLAr and EMLAr CaloCells from CaloUtils/CaloEvent
packages) and/or from parametrization as function of (eta,p)
 refit at vertex
MuidComb:



17/07/2015

gets MuidTracks from previous step and ID reconstructed tracks (iPat)
Muon/ID tracks matches with a c2 cut-off
Gabriella Cataldi INFN Lecce
Combined fit
MuidStandAlone in the HLT framework
Time performance
•MuidStandAlone is executed in a
sequence after the Moore
Algorithms from MooHLT
<SEQUENCE level= ``EF’’
input="LVL1MU"
algorithm="MooHLTAlgo/MooH
LTEFAlgo/m1
MuidHLTAlgo/MuidHLTEFAlgo/
m1"
output="mu" />
PIII 800 MHz 256 MB
6.0.1 rel. - opt build
Average on 1000 events
Pt (GeV)
Time (ms)
20
5.1
100
6.3
•Seeded with Moore tracks
300
•Energy loss from parametrization
•Still needed:
•Calo Reco in HLT for
measured energy loss
H->4mu
•iPat in HLT for track
17/07/2015
Gabriella Cataldi INFN Lecce
mH= 130 GeV
combination
4.9
25.2
Moore/TrigMoore Structure
MooAlgs
MooStatistics
MooMakePhiSegments
MooMakeCrudeRZSegments
MooMakeRoads
MooMakeTracks
…
TrigMoore
MooHLTAlgs
MooMakePhiSegmentsSeeded
MooMakeCrudeRZSegmentsSeeded
MooEvent
17/07/2015
Gabriella Cataldi INFN Lecce
Status of TrigMoore
 Wrapper code so that MOORE can be called by the HLT steering in
unseeded or seeded version IN PLACE
 Seeded code IN PLACE:
 MooMakePhiSegmentSeeded
 MooMakeCrudeRZSegmentsSeeded
Code in place since almost two weeks
Accessing informations about LVL1 and using
the region selector
17/07/2015
Gabriella Cataldi INFN Lecce
Debugging phase




Mistery crash at end of run for all
the muon chain (solved but not
understood)
MapBuilder and RegionSelector
(maps and compact border
problems) (work around / almost
solved) affecting Moore/Mufast
Crash in MuonDetDescr (work
around) no reasons for crashing at
that point! (is the ``not understood
problem coming back’’?)
Time performances in seeded
version. Region selector crash in
optimized mode.
17/07/2015
How long it will take?
Gabriella Cataldi INFN Lecce
Fully wrapped version


Given the status today we only can consider
performances in fully wrapped mode.
This is not very different from running offline.
17/07/2015
Gabriella Cataldi INFN Lecce
First results using Moore in HLT
 A first test has been performed using:
 lxplus036 – Pentium III 800 MHz 256 MB
 opt build
 New EDM
 Average on 500 events
MooAlgs
MooSummary
MooMakePhiSegments
MooMakeCrudeRZSegments
MooMakeRoads
MooMakeTracks
The Athena Chrono service has been used
17/07/2015
Gabriella Cataldi INFN Lecce
TrigMoore – First time-performance test
300GeV
TDR
200GeV
DC1
H
DC1
142 msec
368 msec
279 msec
572 msec
PT
(t -1)
155 msec
• t-1
Average execution time per event calculated for the 500 events sample.
The 1st event has not been included in the calculation since in this event several
services are initialized (magnetic field map,… ).
17/07/2015
4m
20GeV TDR 20GeV DC1
Pt (GeV)
Time (ms)
20
5.1
100
6.3
300
4.9
H->4mu
Gabriella Cataldi INFN Lecce m = 130 GeV
H
25.2
Single m performances

Efficiency vs pT
Moore PT resolution rather
uniform ~ 3% from 6 to 100
GeV
Moore/MuonID performances shown
here are obtained with
• Release 6.0.3
• A private improved version of
MuonIdentification
Tracking in the magnetic
fields
Bug fixes
• Moore with the full material
description
MooAlgs-00-00-41
MooEvent-00-00-42
• Single muons - data set (datasets
0031xx) – 9000 events per file
17/07/2015
With Inert Material Parameterization
And recent improvements in MuonID
PT /GeV
Rather good agreement with
Physics TDR results
Gabriella Cataldi INFN Lecce
1/Pt Resolution vs Pt
Rather good agreement
with Physics TDR
results
Pt /GeV
17/07/2015
Gabriella Cataldi INFN Lecce
Conclusions
Lot’s of problems in this phase for the seeded
version. We (and not only Moore group) are
working very hard to get things ready.
 3 scenarios regarding Moore:
Bugs solved Bugs unsolved Bugs unsolved

Fully wrapped
version seeded
Attempts for an
internal seeding
Ntuple production
17/07/2015
Gabriella Cataldi INFN Lecce
Fully wrapped
version unseeded
“You may say I'm a dreamer,
but…”
J. Lennon