CDF Offline Status and Plans
description
Transcript of CDF Offline Status and Plans
IFC 10/2011 Ray Culbertson1
CDF Offline Status and Plans
Ray Culbertson for the Offline Group
Ray Culbertson, co-leaderAidan Robson, co-leader
Elena Gerchtein, Assoc Head for ProductionStephan Lammel, Assoc. Head for Services
ProductionElena Gerchtein
CalibrationWillis Sukamoto (L)Dan Goldincalibrators
ProductionElena Gerchtein (L)operators
Ntuplingoperators
Monte CarloCostas Vellidis (L)Oksana TadevosyanLiqun LiMC Reps
ServicesStephan Lammel (Mike Kirby)
Grid/DHJoe Boyd (L,CAF)Eric Wicklund (L,DH)Mike Wang(Robert Illingworth)(Marc Mengel)(REX)(Site Coordinators)
DatabasesEric Wicklund (L)(Barry Blumenfeld)(Dennis Box)(DBAs and admins)
Code ManagementJim Bellinger (L)Lynn Garren(Donatella Torretta)
InfrastructureStephan Lammel (L)(CD)
EndingStarting
9/28/2011
Remote SitesAidan Robson (L)Ray Culbertson (L)Site Coordinators
EndingStarting
9/28/2011
CalibratorsSVX Align Dominik HornSVX Timo Aaltonen COT Kevin Burkett dE/dx Keith MateraBeamlines Roberto Carosi TOF Jesus Manuel Vizan GarciaPES Halley BrownPEM Willis Sakumoto,CHA Fabio HappacherCEM Larry NodulmanCal Timing Adam Aurisano,CP2/CCR Azeddine Kasmi PASS Tom Riddick
OperatorsOlga TerlygaOksana TadevosyanZhenbin WuJon WilsonAristotle Calamba
MC RepsHDG Shalhout ShalhoutTOP Dave MietlickiEWK Maria D'ErricoBOT Hideki Miyaki EXO John StrologasQCD Erik Jens Brucken
Site CoordinatorsFermigrid Steve TimmLCG/CNAF Donatella Lucchesi Silvia Amerio Matteo BaucePACAF Masakazu Kurata Yuji Takeuchi Suen Hou Tsan HsiehKISTI Seo-Young Noh Beob Kyun Kim Heejun Yoon Christophe BonnaudMIT Max Goncharov
SAM ShiftersIvan VilaGiovanni Piacentino Stefano Giagu Barry BlumenfeldPeter BusseyThomas Kuhr Alberto Ruiz Aidan Robson
Ntuple maintainersTopntuple Hyunsu LeeBStntuple Michael Morello Angelo Di CantoStntuple Ray Culbertson
IFC 10/2011 Ray Culbertson4
Major Systems Overview
enstore(tape)
productionservers
dCache diskpool Monte Carloservers
ILPProjectDesktops
cdfgrid(5500 slots)
namgridKISTI/MIT/GP/CMS
Off-site farmsEURGRID/PACCAF/
Raw Data
IFC 10/2011 Ray Culbertson5
► Enstore tape system: 16 LTO3 drive, 26 LTO4 drives ○ Delivering typically 20TB per day, 50TB peak► dCache, main disk cache, ~400TB, with tape backend ○ Delivering typically 50TB per day, 150TB peak
► overall very smooth operations last year!
Data Handling
50TB/day >
< one year >
IFC 10/2011 Ray Culbertson6
► dCache major upgrade 300TB → ~800TB in May ○ smooth transition on this major project – great success!
○ appears to have relieved large backlogs - a significant difference!
Data Handling
capacityupgrade
IFC 10/2011 Ray Culbertson7
► Planning needs ○ no new tape drives added last year ○ loads expected to be manageable (thanks to dCache upgrade) ○ new tape storage robot arrived in July, plenty of space ○ purged ~5% in unused datasets
► Tape generation migration ○ from LTO-4 (0.8TB) to T10K (5TB) ○ testing recently signed off ○ 6 T10K drives in FY11, 4 more in FY12 ○ start migrating raw data this year, the bulk of our 9PB in FY12, 13
Data Handling
IFC 10/2011 Ray Culbertson8
CdfGrid
► 5500 slots ► Smooth operations! ► last maintenance replacements purchased in FY11, start decreasing in FY13
► Loads: - heavy but manageable Jan-Jul
40K >
5K >
< 1 year >
IFC 10/2011 Ray Culbertson9
NamGrid
1000 >
< 1 year >
► A portal to offsite farms running OSG, ~20% of CdfGrid ○ achieved regular access to GP and CMS farms! ○ last fall achieved solid integration of the KISTI site! ○ accessing SAM cache and CdfCode ○ regularly runs MC
► MIT site continues to bevery reliable!
►moderate loads
IFC 10/2011 Ray Culbertson10
► Italian colleages reorganized the CNAF and LCGCAF farms ○
Eurogrid
CNAFhead-node
LCGTier1 Farm at CNAF
VO Frontend
GlideIn Factory
glideinWMS
Site CountryCNAF-T1 Italy
INFN-Padova ItalyINFN-Bari Italy
INFN-Legnaro ItalyINFN-Roma1 ItalyINFN-Roma2 ItalyINFN-Catania Italy
INFN-Pisa ItalyFZK-LCG2 GermanyIN2P3-CC France
IFAE SpainPIC Spain
IFC 10/2011 Ray Culbertson11
► the glidein layer prevents LCGCAF from swallowing jobs this makes a huge difference!!► data transfer are also faster due to general network improvement► A huge success! Users are voting for it!► European CDF resources which were languishing are now used!
Eurogrid
< 3 months >
1.5K >
IFC 10/2011 Ray Culbertson12
Diskpool
► 325TB of “persistent” dCache assigned to physics groups► mostly smooth operations► Alexei Varganov, our diskpool expert, has left for a new job► Physics groups have decided the diskpool is less critical now, and we can backup data and live with the greater uncertainty►backup process is very tedious – thanks to production group for their persistence – it will be done soon!
GB statustop 63 85% backed up, progressinghdg 33 done, almost signed offewk 31 down to 3 usersbnt 26 almost doneexo 17 doneqcd 16 investigating
A lot of work:~2000 datasets!
IFC 10/2011 Ray Culbertson13
Code Management
► Last winter developed large file (2 to 8GB) support ○ contributes to improved tape access speed done and deployed!
► New major project: develop legacy code releases ○ incorporate accumulated patches ○ modernize all support packages ○ finally migrate to root version 5! ○ improve infrastructure
first test release is out now, hope to be done this fall!
IFC 10/2011 Ray Culbertson14
Production Operations
► smooth operations over the year
►New data Production ○ 3 billion events, 450 TB
► Ntupling, 3 flavors, data and MC ○ 6 billion events data/MC, 300TB
► Monte Carlo operations ○ Generated 890M events, 150TB last year - continued strong demand!
► Reprocessing…
IFC 10/2011 Ray Culbertson15
Reprocessing
► About half of our data has non-optimal Si clustering, reduces tagging 5-10% (Periods 18-28 out of 38)► Rerun production and ntupling to recover efficiency
we met the Higgs group request for Mar 11 delivery of their data!!
► now only a tiny bit left to do…
IFC 10/2011 Ray Culbertson16
Two More Projects
Two more projects targeted for B physics…
► BStntuple ○ re-ntuple all the B physics data streams ○ add covariance matrix for tracks - allows arbitrary vertexing choices at the ntuple level ○ enables many new analyses, is flexible for the future, and will replace several custom ntuples ► Generic B MC ○ generate, simulate, produce, ntuple bb Monte Carlo ○ has been a wanted for a long, now becoming more urgent ○ details and targeted dataset size are under design
IFC 10/2011 Ray Culbertson17
The Final Push
planning to get the last data out to physics fast !!
P37 prd
P38 calibP38 prd
P38 ntp
P37 ntp
s-scan s-scan
diskpool upload
BStn reprocessing
reprocessing
Sep Oct Nov Dec
Special low energy data!
IFC 10/2011 Ray Culbertson18
In the Long Term
► FY12: cdfgrid, dCache, services continue as-is► FY13: cdfgrid, dCache reduced by size of production needs► Continue with full functionality, reduced capacity for 5y - farm and interactive CPU - access to all data - production and ntupling capability - full MC simulation, with all generators► Continuing past 5y ○ concepts under discussion ○ how will LHC discoveries overlap Tevatron data? ○ will require funding and attracting experts
IFC 10/2011 Ray Culbertson19
The Bottom Line
► CDF Offline had another very successful year! ○ smooth operations, manageable loads ○ clearing off diskpool ○ preparing future releases ○ delivered reproduced data
► As we dive into the new era ○ finishing strong … and fast! ○ long term still requires work, resources Thanks for your IFC contribution, it is crucial!