- genevb's home page
- Posts
- 2024
- 2023
- 2022
- September (1)
- 2021
- 2020
- 2019
- 2018
- 2017
- December (1)
- October (3)
- September (1)
- August (1)
- July (2)
- June (2)
- April (2)
- March (2)
- February (1)
- 2016
- November (2)
- September (1)
- August (2)
- July (1)
- June (2)
- May (2)
- April (1)
- March (5)
- February (2)
- January (1)
- 2015
- December (1)
- October (1)
- September (2)
- June (1)
- May (2)
- April (2)
- March (3)
- February (1)
- January (3)
- 2014
- 2013
- 2012
- 2011
- January (3)
- 2010
- February (4)
- 2009
- 2008
- 2005
- October (1)
- My blog
- Post new blog entry
- All blogs
Sti comparison between SL16b for focus group mods
Updated on Thu, 2016-05-05 23:37. Originally created by genevb on 2016-05-05 23:36.
Several modifications for Sti are being proposed by a focus group. Dmitri S. helped assemble StRoot/Sti in SL16b to include modifications 1-5 from You do not have access to view this node.
I ran two jobs using (1) SL16b as is, and (2) SL16b plus the modifications, on the same Run 13 pp500 dataset I used for previous comparisons [see 1, 2], although this time using the PPV-Boost vertex-finder parameters. I used non-optimized to avoid the known problems with optimized SL16b. The two jobs were run at nearly exactly the same time on an lightly loaded RCF node, allowing for a reasonable timing comparison.
Findings:
1) The modifications had absolutely no effect on track reconstruction.
2) The modifications appear to have slowed down non-optimized Sti by what I estimate to be ~10% ± 1% (stat.) ± 2% (syst.)
Evidence for the latter is demonstrated by using the "Cpu =" numbers from the end of the log files. I get the below ratios of CPU time spent on each timing listed in the logs (modified SL16b ["mod"] divided by standard SL16b ["std"]), where the x-axis in the first plot is an index of the sequence in which timings are listed in the log file (1=bfc, 2=rts_reader, ...16 =Sti, ... etc.). In the second plot, the x-axis is the average of time spent on that item per event, which is relevant to the statistical weight of each value (500 events; I've excluded the full bfc from this plot, which was ~50 CPU-sec/event).
Sti, which accounts for over 50% of the reconstruction time, is ~10% slower in the modified SL16b. Statistical fluctuations in timing have some correlation with the magnitude of the time spent on an item (i.e. if the time is sub-1-second, variations are amplified in the ratio), and I think it's reasonable to estimate that the statistical fluctuations would be at the 1% level or less for Sti, which accounts for nearly 27 CPU-sec/event. I subjectively estimate the systematic error at 2% based on what appears to be not-so-random fluctuations among the non-Sti items. Consequentially, since Sti accounts for just over half the bfc time, the full bfc timing was ~5% slower with the modified SL16b.
-Gene
__________________
For documentation sake, including line numbers corresponding to "index" in the above plots, here are the relevant portions of the logs for the values use...
Standard SL16b:
Modified SL16b:
I ran two jobs using (1) SL16b as is, and (2) SL16b plus the modifications, on the same Run 13 pp500 dataset I used for previous comparisons [see 1, 2], although this time using the PPV-Boost vertex-finder parameters. I used non-optimized to avoid the known problems with optimized SL16b. The two jobs were run at nearly exactly the same time on an lightly loaded RCF node, allowing for a reasonable timing comparison.
Findings:
1) The modifications had absolutely no effect on track reconstruction.
2) The modifications appear to have slowed down non-optimized Sti by what I estimate to be ~10% ± 1% (stat.) ± 2% (syst.)
Evidence for the latter is demonstrated by using the "Cpu =" numbers from the end of the log files. I get the below ratios of CPU time spent on each timing listed in the logs (modified SL16b ["mod"] divided by standard SL16b ["std"]), where the x-axis in the first plot is an index of the sequence in which timings are listed in the log file (1=bfc, 2=rts_reader, ...16 =Sti, ... etc.). In the second plot, the x-axis is the average of time spent on that item per event, which is relevant to the statistical weight of each value (500 events; I've excluded the full bfc from this plot, which was ~50 CPU-sec/event).
Sti, which accounts for over 50% of the reconstruction time, is ~10% slower in the modified SL16b. Statistical fluctuations in timing have some correlation with the magnitude of the time spent on an item (i.e. if the time is sub-1-second, variations are amplified in the ratio), and I think it's reasonable to estimate that the statistical fluctuations would be at the 1% level or less for Sti, which accounts for nearly 27 CPU-sec/event. I subjectively estimate the systematic error at 2% based on what appears to be not-so-random fluctuations among the non-Sti items. Consequentially, since Sti accounts for just over half the bfc time, the full bfc timing was ~5% slower with the modified SL16b.
-Gene
__________________
For documentation sake, including line numbers corresponding to "index" in the above plots, here are the relevant portions of the logs for the values use...
Standard SL16b:
1 QA :INFO - QAInfo:Chain StBFChain::bfc Ast =24961.38 Cpu =24861.82 2 QA :INFO - QAInfo:Maker StRtsReaderMaker::rts_reader Ast = 0.03( 0.0%) Cpu = 0.02( 0.0%) 3 QA :INFO - QAInfo:Maker StAnalysisMaker::analysis Ast = 6.73( 0.0%) Cpu = 6.87( 0.0%) 4 QA :INFO - QAInfo:Maker StTreeMaker::outputStream Ast =692.12( 2.8%) Cpu =656.64( 2.6%) 5 QA :INFO - QAInfo:Maker StEventQAMaker::EventQA Ast = 82.92( 0.3%) Cpu = 82.63( 0.3%) 6 QA :INFO - QAInfo:Maker StMuDstMaker::MuDst Ast =893.45( 3.6%) Cpu =889.51( 3.6%) 7 QA :INFO - QAInfo:Maker StStrangeMuDstMaker::strangeMuDst Ast = 1.97( 0.0%) Cpu = 1.79( 0.0%) 8 QA :INFO - QAInfo:Maker StTagsMaker::tags Ast = 0.78( 0.0%) Cpu = 0.81( 0.0%) 9 QA :INFO - QAInfo:Maker StHighPtTagsMaker::HighPtTags Ast = 0.53( 0.0%) Cpu = 0.40( 0.0%) 10 QA :INFO - QAInfo:Maker StHeavyTagMaker::HeavyTag Ast = 0.40( 0.0%) Cpu = 0.33( 0.0%) 11 QA :INFO - QAInfo:Maker StEventCompendiumMaker::StEventCompendiumMaker Ast = 1.01( 0.0%) Cpu = 1.07( 0.0%) 12 QA :INFO - QAInfo:Maker StSpaceChargeEbyEMaker::scscalercal Ast = 20.15( 0.1%) Cpu = 20.05( 0.1%) 13 QA :INFO - QAInfo:Maker StBTofCalibMaker::btofCalib Ast = 0.47( 0.0%) Cpu = 0.41( 0.0%) 14 QA :INFO - QAInfo:Maker StBTofMatchMaker::btofMatch Ast =109.89( 0.4%) Cpu =109.63( 0.4%) 15 QA :INFO - QAInfo:Maker StdEdxY2Maker::dEdxY2 Ast =5751.06(23.0%) Cpu =5739.94(23.1%) 16 QA :INFO - QAInfo:Maker StiMaker::Sti Ast =13341.69(53.4%) Cpu =13315.69(53.6%) 17 QA :INFO - QAInfo:Maker StGenericVertexMaker::GenericVertex Ast = 0.50( 0.0%) Cpu = 0.41( 0.0%) 18 QA :INFO - QAInfo:Maker StMaker::BTofChain Ast = 19.22( 0.1%) Cpu = 9.28( 0.0%) 19 QA :INFO - QAInfo:Maker StEpcMaker::epc Ast = 0.54( 0.0%) Cpu = 0.57( 0.0%) 20 QA :INFO - QAInfo:Maker StPreEclMaker::preecl Ast = 1.17( 0.0%) Cpu = 1.30( 0.0%) 21 QA :INFO - QAInfo:Maker StEmcRawMaker::emcRaw Ast = 15.59( 0.1%) Cpu = 13.61( 0.1%) 22 QA :INFO - QAInfo:Maker StZdcVertexMaker::StZdcVertexMaker Ast = 0.48( 0.0%) Cpu = 0.60( 0.0%) 23 QA :INFO - QAInfo:Maker StMaker::tpcChain Ast =3987.29(16.0%) Cpu =3979.14(16.0%) 24 QA :INFO - QAInfo:Maker StEventMaker::0Event Ast = 21.83( 0.1%) Cpu = 21.17( 0.1%) 25 QA :INFO - QAInfo:Maker StTriggerDataMaker::trgd Ast = 1.12( 0.0%) Cpu = 1.12( 0.0%) 26 QA :INFO - QAInfo:Maker St_geant_Maker::geant Ast = 1.04( 0.0%) Cpu = 0.98( 0.0%) 27 QA :INFO - QAInfo:Maker StEEmcDbMaker::eeDb Ast = 0.79( 0.0%) Cpu = 0.81( 0.0%) 28 QA :INFO - QAInfo:Maker StTpcDbMaker::tpcDB Ast = 0.92( 0.0%) Cpu = 0.74( 0.0%) 29 QA :INFO - QAInfo:Maker StDetectorDbMaker::detDb Ast = 0.82( 0.0%) Cpu = 1.05( 0.0%) 30 QA :INFO - QAInfo:Maker StMagFMaker::MagField Ast = 0.78( 0.0%) Cpu = 0.73( 0.0%) 31 QA :INFO - QAInfo:Maker St_db_Maker::db Ast = 3.75( 0.0%) Cpu = 3.04( 0.0%) 32 QA :INFO - QAInfo:Maker StIOMaker::inputStream Ast = 2.36( 0.0%) Cpu = 1.48( 0.0%) 33 QA :INFO - QAInfo:Chain StMaker::BTofChain Ast = 5.52 Cpu = 5.49 34 QA :INFO - QAInfo:Maker StVpdCalibMaker::vpdCalib Ast = 0.51( 9.3%) Cpu = 0.54( 9.8%) 35 QA :INFO - QAInfo:Maker StBTofHitMaker::tof_raw Ast = 5.01(90.7%) Cpu = 4.95(90.2%) 36 QA :INFO - QAInfo:Chain StMaker::tpcChain Ast =3986.03 Cpu =3978.22 37 QA :INFO - QAInfo:Maker StTpcHitMover::tpc_hit_mover Ast =3094.64(77.6%) Cpu =3088.68(77.6%) 38 QA :INFO - QAInfo:Maker StTpcHitMaker::tpc_hits Ast =891.38(22.4%) Cpu =889.54(22.4%) 39 QA :INFO - QAInfo:Chain StIOMaker::inputStream Ast = 0.00 Cpu = 1.00 40 QA :INFO - QAInfo:Maker StDAQMaker::inputStream_DAQ Ast = 0.00(100.0%) Cpu = 0.00( 0.0%)
Modified SL16b:
1 QA :INFO - QAInfo:Chain StBFChain::bfc Ast =26269.17 Cpu =26168.58 2 QA :INFO - QAInfo:Maker StRtsReaderMaker::rts_reader Ast = 0.03( 0.0%) Cpu = 0.03( 0.0%) 3 QA :INFO - QAInfo:Maker StAnalysisMaker::analysis Ast = 6.67( 0.0%) Cpu = 6.60( 0.0%) 4 QA :INFO - QAInfo:Maker StTreeMaker::outputStream Ast =705.26( 2.7%) Cpu =667.85( 2.6%) 5 QA :INFO - QAInfo:Maker StEventQAMaker::EventQA Ast = 78.44( 0.3%) Cpu = 78.25( 0.3%) 6 QA :INFO - QAInfo:Maker StMuDstMaker::MuDst Ast =888.64( 3.4%) Cpu =885.98( 3.4%) 7 QA :INFO - QAInfo:Maker StStrangeMuDstMaker::strangeMuDst Ast = 1.87( 0.0%) Cpu = 2.13( 0.0%) 8 QA :INFO - QAInfo:Maker StTagsMaker::tags Ast = 0.75( 0.0%) Cpu = 0.65( 0.0%) 9 QA :INFO - QAInfo:Maker StHighPtTagsMaker::HighPtTags Ast = 0.50( 0.0%) Cpu = 0.46( 0.0%) 10 QA :INFO - QAInfo:Maker StHeavyTagMaker::HeavyTag Ast = 0.37( 0.0%) Cpu = 0.34( 0.0%) 11 QA :INFO - QAInfo:Maker StEventCompendiumMaker::StEventCompendiumMaker Ast = 0.97( 0.0%) Cpu = 0.94( 0.0%) 12 QA :INFO - QAInfo:Maker StSpaceChargeEbyEMaker::scscalercal Ast = 20.12( 0.1%) Cpu = 20.02( 0.1%) 13 QA :INFO - QAInfo:Maker StBTofCalibMaker::btofCalib Ast = 0.43( 0.0%) Cpu = 0.55( 0.0%) 14 QA :INFO - QAInfo:Maker StBTofMatchMaker::btofMatch Ast =108.56( 0.4%) Cpu =108.18( 0.4%) 15 QA :INFO - QAInfo:Maker StdEdxY2Maker::dEdxY2 Ast =5825.40(22.2%) Cpu =5814.03(22.2%) 16 QA :INFO - QAInfo:Maker StiMaker::Sti Ast =14695.96(55.9%) Cpu =14667.24(56.0%) 17 QA :INFO - QAInfo:Maker StGenericVertexMaker::GenericVertex Ast = 0.45( 0.0%) Cpu = 0.47( 0.0%) 18 QA :INFO - QAInfo:Maker StMaker::BTofChain Ast = 16.63( 0.1%) Cpu = 9.97( 0.0%) 19 QA :INFO - QAInfo:Maker StEpcMaker::epc Ast = 0.50( 0.0%) Cpu = 0.38( 0.0%) 20 QA :INFO - QAInfo:Maker StPreEclMaker::preecl Ast = 1.12( 0.0%) Cpu = 1.01( 0.0%) 21 QA :INFO - QAInfo:Maker StEmcRawMaker::emcRaw Ast = 15.60( 0.1%) Cpu = 14.20( 0.1%) 22 QA :INFO - QAInfo:Maker StZdcVertexMaker::StZdcVertexMaker Ast = 0.44( 0.0%) Cpu = 0.36( 0.0%) 23 QA :INFO - QAInfo:Maker StMaker::tpcChain Ast =3865.29(14.7%) Cpu =3857.50(14.7%) 24 QA :INFO - QAInfo:Maker StEventMaker::0Event Ast = 22.69( 0.1%) Cpu = 22.31( 0.1%) 25 QA :INFO - QAInfo:Maker StTriggerDataMaker::trgd Ast = 1.02( 0.0%) Cpu = 1.05( 0.0%) 26 QA :INFO - QAInfo:Maker St_geant_Maker::geant Ast = 2.47( 0.0%) Cpu = 0.99( 0.0%) 27 QA :INFO - QAInfo:Maker StEEmcDbMaker::eeDb Ast = 0.70( 0.0%) Cpu = 0.61( 0.0%) 28 QA :INFO - QAInfo:Maker StTpcDbMaker::tpcDB Ast = 0.84( 0.0%) Cpu = 0.96( 0.0%) 29 QA :INFO - QAInfo:Maker StDetectorDbMaker::detDb Ast = 0.74( 0.0%) Cpu = 0.68( 0.0%) 30 QA :INFO - QAInfo:Maker StMagFMaker::MagField Ast = 0.70( 0.0%) Cpu = 0.65( 0.0%) 31 QA :INFO - QAInfo:Maker St_db_Maker::db Ast = 4.47( 0.0%) Cpu = 2.64( 0.0%) 32 QA :INFO - QAInfo:Maker StIOMaker::inputStream Ast = 1.52( 0.0%) Cpu = 1.55( 0.0%) 33 QA :INFO - QAInfo:Chain StMaker::BTofChain Ast = 5.39 Cpu = 5.51 34 QA :INFO - QAInfo:Maker StVpdCalibMaker::vpdCalib Ast = 0.47( 8.8%) Cpu = 0.44( 8.0%) 35 QA :INFO - QAInfo:Maker StBTofHitMaker::tof_raw Ast = 4.91(91.2%) Cpu = 5.07(92.0%) 36 QA :INFO - QAInfo:Chain StMaker::tpcChain Ast =3864.21 Cpu =3856.60 37 QA :INFO - QAInfo:Maker StTpcHitMover::tpc_hit_mover Ast =2940.59(76.1%) Cpu =2934.90(76.1%) 38 QA :INFO - QAInfo:Maker StTpcHitMaker::tpc_hits Ast =923.62(23.9%) Cpu =921.70(23.9%) 39 QA :INFO - QAInfo:Chain StIOMaker::inputStream Ast = 0.00 Cpu = 1.00 40 QA :INFO - QAInfo:Maker StDAQMaker::inputStream_DAQ Ast = 0.00(100.0%) Cpu = 0.00( 0.0%)
»
- genevb's blog
- Login or register to post comments