Direct use of IMagFieldSvc in SiDetElementsRoadMaker_xk (ATLASRECTS-4728)

Full details available at NICOS MR-20629-2019-01-25-13-49
Athena: number of compilation errors 0, warnings 0
AthSimulation: number of compilation errors 0, warnings 0
CI Jenkins server is switched to https://atlas-sit-ci.cern.ch. It is accessible world-wide (behind CERN SSO). In old links to Jenkins server aibuild080.cern.ch:8080 should be replaced with atlas-sit-ci.cern.ch For experts only: Jenkins output [CI-MERGE-REQUEST 32755]

Hi @goetz

I modified SiDetElementsRoadMaker_xk to directly access AtlasFieldSvc instead of callback. I ran q221 and q431 tests at lxplus050 with Athena master x86_64-slc6-gcc8-opt 2019-01-21T2352.

In RDOtoRDOTrigger of q221, one event took 3.08 s and the increase due to the change was 4.50 ms. This corresponds to 0.15%.
In RAWtoESD of q221, one event took 5.56 s and the increase due to the change was 1.61 ms. This corresponds to 0.03%.
In RAWtoESD of q431, one event took 14.2 s and the increase due to the change was 28.3 ms. This corresponds to 0.20%.

The increase is not large.

Cheers, Susumu

Maybe we should discuss this in the next reco meeting. I would tend towards using the field service directly and just leaving a comment that this may require a re-optimisation if this turns out to be too slow.

In principle it would be good to test this on a "high" pileup sample. But I would not know a good example job. The "best" I can imagine is the ttbar art test : (https://gitlab.cern.ch/atlas/athena/blob/master/InnerDetector/InDetValidation/InDetPerformanceRTT/test/test_InDetPerformanceRTT_ttbar.sh#L1). Could You eventually also check the performance for this job with e.g. 100 events or so (If You have a pile up test job that would be even better) ?

Thanks ! Götz

Hi @goetz,

I ran a MC16e style job at lxplus109.cern.ch

Reco_tf.py --digiSteeringConf 'StandardSignalOnlyTruth' --conditionsTag 'default:OFLCOND-MC16-SDR-25' --valid 'True' --pileupFinalBunch '6' --numberOfHighPtMinBias '0.2595392' --autoConfiguration 'everything' --numberOfLowPtMinBias '99.2404608' \
--preInclude 'HITtoRDO:Digitization/ForceUseOfPileUpTools.py,SimulationJobOptions/preInlcude.PileUpBunchTrainsMC16c_2017_Config1.py,RunDependentSimData/configLumi_run310000.py' --postExec 'all:CfgMgr.MessageSvc().setError+=["HepMcParticleLink"]' "ESDtoAOD:fixedAttrib=[s if 'CONTAINER_SPLITLEVEL = \'99\'' not in s else '' for s in svcMgr.AthenaPoolCnvSvc.PoolAttributes];svcMgr.AthenaPoolCnvSvc.PoolAttributes=fixedAttrib" --postInclude 'default:PyJobTransforms/UseFrontier.py' --preExec 'all:rec.Commissioning.set_Value_and_Lock(True);from AthenaCommon.BeamFlags import jobproperties;jobproperties.Beam.numberOfCollisions.set_Value_and_Lock(20.0);from LArROD.LArRODFlags import larRODFlags;larRODFlags.NumberOfCollisions.set_Value_and_Lock(20);larRODFlags.nSamples.set_Value_and_Lock(4);larRODFlags.doOFCPileupOptimization.set_Value_and_Lock(True);larRODFlags.firstSample.set_Value_and_Lock(0);larRODFlags.useHighestGainAutoCorr.set_Value_and_Lock(True); from LArDigitization.LArDigitizationFlags import jobproperties;jobproperties.LArDigitizationFlags.useEmecIwHighGain.set_Value_and_Lock(False)' 'ESDtoAOD:from TriggerJobOpts.TriggerFlags import TriggerFlags;TriggerFlags.AODEDMSet.set_Value_and_Lock("AODSLIM");' \
--geometryVersion 'default:ATLAS-R2-2016-01-00-01' --numberOfCavernBkg '0'  --inputLowPtMinbiasHitsFile mc16_13TeV.361238.Pythia8EvtGen_A3NNPDF23LO_minbias_inelastic_low.simul.HITS.e4981_s3087_s3111/HITS.10701323._007457.pool.root.1 --inputHighPtMinbiasHitsFile mc16_13TeV.361239.Pythia8EvtGen_A3NNPDF23LO_minbias_inelastic_high.simul.HITS.e4981_s3087_s3111/HITS.10701335._002690.pool.root.1 --inputHITSFile mc16_13TeV.410470.PhPy8EG_A14_ttbar_hdamp258p75_nonallhad.simul.HITS.e6337_s3126/HITS.12458448._055308.pool.root.1 --outputESDFile ESD.pool.root --maxEvents 100 --digiSeedOffset1 1 --digiSeedOffset2 1 --jobNumber 1 \
--steering 'doRDO_TRIG'

with Athena master 2019-01-26T2139 x86_64-centos7-gcc8-opt with the change. From the Reco_tf.py command, I dropped --triggerConfig 'RDOtoRDOTrigger=MCRECO:DBF:TRIGGERDBMC:2216,76,260' because of crash.

In RAWtoESD, 36217.778 ms took for one event, 21.10064245 ms was increased. This corresponds to 0.06%.
In RDOtoRDOTrigger, 6677.576 ms took for one event, 4.52763480 ms was increased. This corresponds to 0.07%.

The pileup distribution of the processed 100 events is .

added 1 commit

aa005ac9 - Update

Compare with previous version

Thanks, Susumu !

I was wondering what the impact would be if the z-field would be computed using the Tp.position() rather than the centre (may have some negative impact on performance since the caching might not work as well I guess). One could also try to get rid of the string comparison and define already during initialize fieldModeEnum (which is then only changed to Trk::NoField if the solenoid is off).

added 197 commits

aa005ac9...a9f60e61 - 196 commits from branch atlas:master
cc06665b - Merge remote-tracking branch 'upstream/master' into master-ATLASRECTS-4728-v2

Compare with previous version

added 1 commit

8aa3b4b2 - Avoid string comparison

Compare with previous version

added 1 commit

569ec98d - Use position from TrackParameter

Compare with previous version

Hi @goetz,

I modified the code to avoid string comparison and to use the position from the track parameter.

In RAWtoESD, 36217.778 ms took for one event, 10.409295 ms was increased. This corresponds to 0.03%.
In RDOtoRDOTrigger, 7006.263 ms took for one event, 2.49427015 ms was increased. This corresponds to 0.04%.

Cheers, Susumu

Direct use of IMagFieldSvc in SiDetElementsRoadMaker_xk (ATLASRECTS-4728)

Merged by Walter Lampl 6 years ago (Feb 6, 2019 1:53pm UTC) 6 years ago

Activity

Direct use of IMagFieldSvc in SiDetElementsRoadMaker_xk (ATLASRECTS-4728)

Merge request reports

Merged by Walter Lampl 6 years ago (Feb 6, 2019 1:53pm UTC) 6 years ago

Activity