CDFII Computing Status Donatella Lucchesi OUTLINE Ø Status

CDFII Computing Status Donatella Lucchesi OUTLINE: Ø Status of production and data processing at FNAL Ø CAF usage and resources for users Ø Status of our CAF farm and usage of TIER 1 Ø New possibilities under discussion 12/09/2005 Donatella Lucchesi 1

Production status ØThe single-pass production processing scheme now in production: § Process stream B G J in real time to make calibrations then process data only once Ø Production Farm: ~20 Mln/day, 15 -20 pb-1 per week Ø Data processed until 19 May 2005 Ø Running production on data taken [20 May - 20 Jul] Ø Calibration finished for data taken [21 Jul – 29 Aug] Ø Data access via SAM, seems stable Ø Small datasets kept on disk Ø Large datasets? 12/09/2005 Donatella Lucchesi 2

CAF status at Fnal: Italian resources (latest 6 months) All slots used Many waiting process 12/09/2005 Donatella Lucchesi 3

CAF status at Fnal (latest 6 months) CAF Condor. CAF Running sessions Waiting sessions Full! 12/09/2005 Donatella Lucchesi 4

Current situation for users ØStart to produce entuples/skimmed data for winter Conferences. ØIssues for the BCHARM dataset (Bs mixing! ) : q Is the cpu on site enough to analyze all the data? q Is the disk space sufficient for data and MC ? ØNew disks pool (~350 TB) available soon: how to manage disk space and user access under review. ØOther possible actions: q make stntuple and skimmed dataset official q store secondary datasets outside Fermilab 12/09/2005 Donatella Lucchesi 5

Status at CNAF: usage of our farm Maintained by Subir Sarkar All the cpus fully occupied and waiting sessions The Italians are the major users 12/09/2005 Donatella Lucchesi 6

Status at CNAF: Glide. CNAF Developed and maintained by Subir Sarkar with Igor Sfiligoi help Other glidecaf: Fermi. Grid-> beta test Planned at: Lyon, Barcelona, Karlsruhe and GLOW (Grid Laboratory of Wisconsin) • CDF way to access Tier 1 resources • Fixed share: 60% Italians 40% Others. Plus a better priority for Italy queue • Italians can submit to common Great success!!! Used to produce Monte Carlo but also to run on data Which data? 12/09/2005 Donatella Lucchesi 7

CNAF status: Data Access and Local Storage Maintained by F. Delli Paoli, A. Fella and D. L. Via SAM access: q Data al fnal: copy via network as job need q Data stored at CNAF Dataset Monte Carlo for top Size (TB) 0. 150 J/ 1. 1 High Pt muon 0. 9 High Pt electron 0. 4 B muon 1 B electron 1. 4 B->hh 0. 7 B CHARM 14 Skimmed data Most important datasets @CNAF. Data accessed by Italians and non using CAF@CNAF and Glide. CNAF 4 12/09/2005 Donatella Lucchesi 8

More resources on GRID: g. Lite CAF Developers: F. Delli Paoli, A. Fella, D. Jeans Other responsibilities: S. Sarkar, I. Sfiligoi and D. L. CDF users submit jobs on the GRID using the same tools they have been using till today. Only for MC production Alfa test in Padova 12/09/2005 Donatella Lucchesi 9

Bottom Lines On April 22 nd: “ ØUsers happy, analysis not limited by resources ØImpossible to define now if and when we will have crisis “ Now: Ø CDF at Fermilab is becoming resources limited Ø B group has the most important impact Ø Solutions: make stntuples and skimmed data “official” Ø Since CNAF is performing so well move secondary dataset and/or tertiary datasets here Ø Need to quantify cpu, disks and human resources Ø Tomorrow starts the Fermilab Computing review, more inputs after that 12/09/2005 Donatella Lucchesi 10

Disk Space Presented on April 22 nd: Ø CDF produces ~100 TB/year ~20% at CNAF 20 TB/year Ø Same amount of disk space considered for Monte Carlo Ntuples and reduced datasets homemade Ø End of 2005 CDF has 90 TB (80 from CNAF) Ø Small expansion: Year TB CDF TB T 1 %CDF +40 TB/year 2004 10 Ø CDF is a small fraction of 2005 32 200 16% CNAF/Tier 1 program 2006 90 850 11% Ø Years of reference: 2008/9 2007 130 1500 9% CDF = 4~5% of Tier 1 12/09/2005 170 3500 5% 2009 210 5600 4% 2010 We may need more disk space 2008 250 9000 3% Donatella Lucchesi 11

CPU Presented on April 22 nd: CDF T 1 Year 2004 2005 (KSI 2 K) 86 494. 5 1300 %CDF 38% 2006 2007 2008 2009 2010 903 1032 1161 1290 1419 1800 2400 6300 10500 18000 50% 43% 18% 12% 8% We may need a bigger share with other experiments 12/09/2005 Donatella Lucchesi 12

People Needs assessment, interface with CSN 1, Tier 1 and GRID, HW resources management in Italy and USA Donatella Lucchesi Francesco Delli Armando Fella Paoli End Oct. 06 End Nov. 05 data replica, User access and support management in Italy Code distribution in Italy Igor Sfiligoi Subir Sarkar End Jul. 06 Large dataset management, split, skim, concatenation CDF-GRID coord. and CAF’s management at FNAL, d. CAF cpus, disks and code coordinator, help in coordinating activities CNAF/FNAL Daniel Jeans (50%) Management and operation of CNAF farm, migration to Tier 1 common pool Porting CAF code to LCG maintain g. Lite CAF We need at least as many people as we are now for maintenance and if we want to support data analysis 12/09/2005 Donatella Lucchesi 13

Backup 12/09/2005 Donatella Lucchesi 14

CPU: “Assisi” CDF requests Ø Scenario presented in 2004 (3 times) based on CDF official expectations presented by Rick Snider and not changed as today Ø CPU for CDF-Italy users: linear increase as 2004 2005 Ø CPU for CDF: 15% of total needs as presented Ø In the reference year 2008/9 CDF ask for CPU as Atlas or CMS Ø Not crazy, but difficult to make now the case Ø Today proposal: we will ask so much when and if we need 12/09/2005 Donatella Lucchesi 15

CPU: “Assisi” CDF requests CDF T 1 Year 2004 2005 (k. SI 2 K) 86 495 1300 %CDF 38% 2006 2007 2008 2009 2010 12/09/2005 (KSI 2 K) 903 1505 2107 2709 3311 1800 2400 6300 10500 18000 50% 63% 33% 26% 18% Donatella Lucchesi 16

CPU: exercise “crescita zero” Ø Assuming, as exercise, that starting from 2006 CDF does not buy any CPU we have 900 KSI 2 K Ø In the reference year 2008/9 CDF is 12% of Tier 1 a bit less of ~1/3 of Atlas o CMS Ø Not surprising, CDF will have ~10 PB of data Ø CDF will not be “insignificant” in any case 12/09/2005 Donatella Lucchesi 17

CPU: exercise “crescita zero” CDF Year 2004 2005 2006 2007 2008 2009 2010 12/09/2005 (KSI 2 K) 903 903 903 86 494. 5 T 1 (k. SI 2 K) 1300 1800 2400 6300 10500 18000 Donatella Lucchesi %CDF 38% 50% 38% 14% 9% 5% 18

CPU: “in medio stat virtus” Ø But, CDF does not disappear Ø A small increase per year is needed: § luminosity will increase and data as well § to have something to count on in case GRID or FNAL are too loaded § give the possibility to redefine each year the increase (positive or negative) Ø Assumption: increase of 300 KSI 2 K/year after 2006 Ø In the reference year 2008/9 CDF is 15% of Tier 1 to be compared with 12% if “a crescita zero” 12/09/2005 Donatella Lucchesi 19

Backup 12/09/2005 Donatella Lucchesi 20

d. CAF @CDF 12/09/2005 Donatella Lucchesi 21

Assisi requests CDF Analysis hardware plan Roadmap for CNAF 12/09/2005 Donatella Lucchesi 22