T2s daily tests

This page contains an example of how CPT can be used for executing daily tests on T2.

Configuration

For these tests the /QCD_Pt80/Summer09-MC_31X_V3_AODSIM-v1/AODSIM has been used, with some cross-check with the /InclusiveBB_Pt30/Summer09-MC_31X_V3_7TeV-v1/GEN-SIM-RECO dataset (the choice was dependent on the data ditribution at the chosen sites).

Three config files has been used:

• A standard PAT-tuplizer, CMSSW_3_3_6 and 50k evts/job http://cmssw.cvs.cern.ch/cgi-bin/cmssw.cgi/CMSSW/PhysicsTools/PatAlgos/test/patLayer1_fromAOD_full_cfg.py
• A JetAnalyzer CMSSW_3_3_6 and 50k evts/job http://cmssw.cvs.cern.ch/cgi-bin/cmssw.cgi/CMSSW/JetMETAnalysis/JetAnalyzers/ (tag TV00-05-05)

a "drop *" has been added as usual

Two different TFileAdaptor have been used:

• No settings
• ReadHint, CacheHint=Auto, cache=20MB

The following sites were involved in the tests:

• CSCS
• KNU
• Nebraska
• Pisa
• Vienna
• Wisconsin

Some thoughts on test configurations here

CSCS

• CSCS-PAT1-QCD_Pt80Summer09.MC_31X_V3_AODSIM.v1AODSIM-Overview.jpg:
  

• CSCS-PAT1_RHAUTO_CHAUTO_CACHE20-QCD_Pt80Summer09.MC_31X_V3_AODSIM.v1AODSIM-Overview.jpg:
  

• CSCS-JRA1-QCD_Pt80Summer09.MC_31X_V3_AODSIM.v1AODSIM-Overview.jpg:
  

• CSCS-JRA1_RHAUTO_CHAUTO_CACHE20-QCD_Pt80Summer09.MC_31X_V3_AODSIM.v1AODSIM-Overview.jpg:
  

KNU

Not big differences during the tests.

• KNU-PAT1-20100202-Overview.jpg:
  

• KNU-PAT1_RHAUTO_CHAUTO_CACHE20-20100209-Overview.jpg:
  

• KNU-JRA1-20100210-Overview.jpg:
  

• KNU-JRA1_RHAUTO_CHAUTO_CACHE20-20100208-Overview.jpg:
  

Nebraska

Nothing to report

• Nebraska-PAT1-20100209-Overview.jpg:
  

• Nebraska-PAT1_RHAUTO_CHAUTO_CACHE20-20100208-Overview.jpg:
  

• Nebraska-JRA1-20100209-Overview.jpg:
  

• Nebraska-JRA1_RHAUTO_CHAUTO_CACHE20-20100208-Overview.jpg:
  

Pisa

No big differences among the various days, some tails on 2010-02-09

A further check has been done comparing:

• QCD_Pt80Summer09.MC_31X_V3_AODSIM.v1
• InclusiveBB_Pt30+Summer09.MC_31X_V3_7TeV.v1-RECO

Where the second dataset has been chosen among the ones available at the site. Despite the fact that the latter is actually reading less data (and so consumes less time), both of them traces well the site performances

• Pisa-PAT1-QCD_Pt80Summer09.MC_31X_V3_AODSIM.v1AODSIM-Overview.jpg:
  

• Pisa-PAT1_RHAUTO_CHAUTO_CACHE20-QCD_Pt80Summer09.MC_31X_V3_AODSIM.v1AODSIM-Overview.jpg:
  

• Pisa-JRA1-QCD_Pt80Summer09.MC_31X_V3_AODSIM.v1AODSIM-Overview.jpg:
  

• Pisa-JRA1_RHAUTO_CHAPP_CACHE20-QCD_Pt80Summer09.MC_31X_V3_AODSIM.v1AODSIM-Overview.jpg:
  

• Pisa-PAT1-QCDvsBB-Overview.jpg:
  

Vienna

Vienna seems to have some instability in the file reading:

• Big failures on (1) dataset
• Reading time varies quite a lot during the testing days

• Vienna-PAT1-Overview.jpg:
  

• Vienna-PAT1_RHAUTO_CHAUTO_CACHE20-Overview.jpg:
  

• Vienna-PAT1-InclusiveBB_Pt30Summer09.MC_31X_V3_7TeV.v1GEN.SIM.RECO-Overview.jpg:
  

• Vienna-JRA1-Overview.jpg:
  

• Vienna-JRA1_RHAUTO_CHAUTO_CACHE20-Overview.jpg:
  

• Vienna-JRA1-InclusiveBB_Pt30Summer09.MC_31X_V3_7TeV.v1GEN.SIM.RECO-Overview.jpg:
  

Wisconsin

Nothing to report

• Wisconsin-PAT1-20100202-Overview.jpg:
  

• Wisconsin-PAT1_RHAUTO_CHAUTO_CACHE20-20100209-Overview.jpg:
  

• Wisconsin-JRA1-20100209-Overview.jpg:
  

• Wisconsin-JRA1_RHAUTO_CHAPP_CACHE20-20100211-Overview.jpg:
  

Some Thoughts

There seem not to exist big differences between running on a AOD or a RECO sample (at a first glance). Both of them seem to be sensitive to performance degradations. In this case, an AOD sample would be lighter to distribute amont T2s.

Both JRA and PAT seem sensitive to performance degratation; anyway, they greatly differ in read data quantity (PAT>JRA) and time/event (PAT:~3h, JRA~30 min). The choice between these two is the choice between a short and light job (faster turnaround, less load on the site but more prone to perturbations) and a longer and heavier one (slow turnaround, more load on site but more stable). My proposal is to use both of them at least at the beginning.

Also, I would stick on Std CMSSW settings, as using the AUTO ones (mostly LazyDownload) will boost too much performances and does not put a measurable load on the data reading part (usually the bottleneck)

-- LeonardoSala - 11-Feb-2010