Welcome to the Computing web
한국 CMS그룹에서 Tier3자원 확충을 위한 task force가 구성되었습니다.
현재의 8TeV analysis와 함께 향후 LHC Run-II를 대비해 필요한 Tier3 자원에 대해 조사하여 이를 확충하는 데 어떠한 것들이 필요한지에 대해 정리합니다.
Common understanding and decisions
- Expected resources dedicated for KCMS T3
- 150TB storage, dedicated to T3
- Upgrade plans (in progress)
- Ntuple data format
- SKNtuple based on LQNtuple (J. Almond)
- Skimmed dataset definition :
- Single lepton : HLT + vertex + Isolated muon + dilepton veto + jet multiplicity -> reduced to 5% level for ttbar MC
- A central skim is available
- Dilepton : HLT + vertex + loose id 2 muons + loose low mass cut -> reduced to 10% level for real data
- Need common skim path for MC
- Comparisons of of currently AvailableResources in tier1,2,3 sites
- T1 average : 16130 HS06 / 7585.5 TB
- T2 average : 8432 HS06 / 603.44 TB
- CERN T3 : 4800 HS06 / 1250TB
- KNU : ~1500 HS06 / 20 nodes(125 core) / 450TB (shared with T2)
- UOS : 723.8 HS06 / 15 nodes(56 core) / 66TB
- Estimation of necessary resources for the physics analysis DesiredResources
- Dileptons AOD : about 100 TB
- Semileptons + Higgs AOD : 277TB
- Semilepton MC is superset of MC for dileptons
- Summing up, we need ~360TB to store AOD. this is too big
Meetings / action items
6th KCMS T3 meeting - 6 Nov 2013
- SNU server plan
- New server with 128 cores will be available for CMS analysis
- LQNtuple and SKNtuple data structure
- 3 days to process and produce ntuples
- Dilepton channel checks
- Dilepton skim status - MuEG? done, ongoing for others
- UoS? upgrade plan - 10G network test next week
5th KCMS T3 meeting - 25 Oct 2013
- 5th KCMS T3 TF meeting :
- Dilepton skimming
- Finalized skim cuts after circulating in forum
- Crab jobs in progress, A student is taking care of job submissions
- Many 60317 failures
- Common configuration is available in github.com, need input from L+J analysis
- Lepton+Jet skimming
- Central skim definitions and corresponding secondary datasets for Top group already exist, asked to analyzers for details
- UoS? T3 operation
- Network usage monitoring
- SE upgrade is planned, ordering optical fiber network - fund from Seoul City
- Introduced batch system, need manpower (MS level) to build job submission interface
- T3 TF activity
- Suyong will visit CERN for RRB meeting. Ask Suyong to put T3 TF efforts into his slides
4th KCMS T3 meeting - 18 Oct 2013
- 4th KCMS T3 TF meeting : https://indico.cern.ch/conferenceDisplay.py?confId=278754
- Dataset skimming
- Idea on common skim workflow for leptonJet + dilepton
- First trial of dilepton skim - 한번 당 500개 job section정도로 submit해야 성공률이 높음
- UoS? load test with skim jobs - 전송 속도는 500Mbps가량
- T3 resource 추가 증설
- 경북대 - T2 site증설 계획안이 있음. KCMST3전용으로 분리시켜 100TB추가 증설 가능하다. CPU자원 사용을 위해 T2에 연동하는 경우 문제가 있을 수 있음.
- 서울시립대 - 사용량을 높여 서울시 등에 예산 추가 신청할 여지가 있음.
- KISTI - 300코어 정도 제공 받을 수 있을 것으로 보임. 단, KCMS T3를 전담하는 인력 배치를 확약받아야 한다.
3rd KCMS T3 TF meeting - 11 Oct 2013
- 3rd KCMS T3 TF meeting : https://indico.cern.ch/conferenceDisplay.py?confId=277560
- Next meeting : 18 Oct 3pm (Friday afternoon)
- Skimmed AOD sample을 정의하고 이를 활용하기로 결정함
- Reduction rate in Single lepton ~ 10% (G. Yu)
- 총합 150TB수준에서 selection cut을 정의할 수 있게 한다.
- Single lepton : G. Ryu / Dilepton : S. Song&J. Goh DataSkim
- Crab timeout 문제에 대한 해결 가능성을 확인함
- 직접 transfer만 따로 한 후 fjr 파일을 수정하거나 cmscp.py를 수정하면 timeout을 조정할 수 있음 (G. Ryu)
- 추가 용량 증설을 계획함
- 향후 500TB, 200 core수준을 희망함.
- KCMS펀드에서 2000만원의 예산을 할당하여 100TB를 증설할 수 있도록 계획 - 경북대 T3에 증설 가능한지 확인 (S. Choi)
- 이후 KISTI에서 추가 resource를 받을 수 있는지 문의 (S. Choi)
2nd KCMS T3 TF meeting - 02 Oct 2013
KCMS T3 TF kickoff meeting - 16 Sep 2013
- Chairperson : 유인태 교수님
- Currently available resources - 류건모 (UOS, KNU, other sites)
- Survey of T3 computing power and storage capacity – T2 급으로
- Operation of T3 – pbs batch script, phedex, rfio, crab(?)
- Data production and analysis pattern – 고정환, 송상현
- Pat-tuple production
- Piggy back option – ex) use B2G? group pat-tuple
- Data sample size – (AOD, signal, background, control)
- Dilepton analysis 위한 data set– 고정환, 송상현
- Lepton + jets 위한 data set – 유금봉
- Lepton + MET – 김동희 교수님