Download presentation
Presentation is loading. Please wait.
Published byGarey Eaton Modified over 8 years ago
1
9 November 98 1 Jürgen Knobloch ATLAS Computing Overview of ATLAS Computing Jürgen Knobloch Slides also on: http://nicewww.cern.ch/atlas/wwwslides/DOE9811/index.htm
2
9 November 98 2 Jürgen Knobloch ATLAS Computing Outline Software –Quality –Strategy –Human resources –Current activities Hardware –Requirements –Regional centers Organizational aspects Conclusions
3
9 November 98 3 Jürgen Knobloch ATLAS Computing We are at a turning point... Physics “TDR” - spring 99 –based principally on FORTRAN software –benchmark for detector and software performance Internal Review of Computing –Committee created by ATLAS management –Lead by Homer Neal –Report by February 99 Preparatory work for OO software is done –GEANT-4 production release - December 98 –LHC++ - “CERNLIB” for OO developments –ARVE - ATLAS framework available –Software process support - Rules, SRT, ASP
4
9 November 98 4 Jürgen Knobloch ATLAS Computing Offline Software Generate Events Generate Events Simulate Events Simulate Events Simulation geometry Build Simulation Geometry Build Simulation Geometry Reconstuction geometry Build Reconstruction Geometry Build Reconstruction Geometry Detector description Detector alignment Detector calibration Reconstruction parameters Reconstruct Events Reconstruct Events ESD AOD Analyze Events Analyze Events Physics Raw Data ATLAS Detector
5
9 November 98 5 Jürgen Knobloch ATLAS Computing Software Quality Functionality, correctness, robustness –Experience, requirements, software process Ease of entry for physicists –Documentation, clear architecture, training, mentors Maintainability –Documentation, software process Flexibility –changing environment, new requirements Performance –Few O(0) re-processings, CPU, memory, I/O
6
9 November 98 6 Jürgen Knobloch ATLAS Computing Software Strategy Defined process - ATLAS Software Process (ASP) –Reviews of deliverables –Tools (SRT, Code checking, LIGHT, GNATS, …) Object oriented design and implementation - C++ Standards and commercial components wherever possible and reasonable Common solutions with other experiments wherever possible –LCB projects GEANT-4, LHC++, RD45, Spider, Monarc, … Learn from other experiments’ experience –BaBar, Fermilab, STAR,...
7
9 November 98 7 Jürgen Knobloch ATLAS Computing Trigger Domain Decomposition GUI Control Event Display Muon System Magnetic Field Calorimetry Detector Description Inner Detector Data Base Simulation Event Tools SDE Documentation Muon Reconst. e, Reconstruction Organisation: Domain architects Domain coordinators Domain interface group Support team Create “centres of gravity” for domains
8
9 November 98 8 Jürgen Knobloch ATLAS Computing Domains and People FTE nowFTE required Management, general support616 –Coordinators, Librarian, Software process support Reconstruction25 Simulation25 Event domain35 Graphics46 Analysis tools13 Field, alignment, calibration14 Detector description13 System Domains:3644 –Trigger Simulation, Inner Detector, Muon Spectrometer, Calorimeters _________________________________________________ SUM5691
9
9 November 98 9 Jürgen Knobloch ATLAS Computing Manpower evolution Re-direct people from physics studies in FORTRAN to developments in OO technology. In addition need software professionals for key tasks: Software process support, framework, data base: ~25 + support of regional centers + R&D ATLAS Computing Workforce Integral 1994-2005: 900 person/years FORTRAN OO
10
9 November 98 10 Jürgen Knobloch ATLAS Computing Timeline - past 1992 | 1993 | 1994 | 1995 | 1996 | 1997 | 1998 Preparatory Phase LOITPGEANT4CTP RD45 1 Tbyte event DB ARVE MOOSE
11
9 November 98 11 Jürgen Knobloch ATLAS Computing Timeline - future 1999 | 2000 | 2001 | 2002 | 2003 | 2004 | 2005 Implementation and Commissioning Phase Physics “TDR”First OO simulation First OO reconstruction Complete OO simulation Complete OO reconstruction Decide on DatabaseFull Test OpSys Operation Computing Technical Proposal - Rel 2
12
9 November 98 12 Jürgen Knobloch ATLAS Computing ATLAS Software Process (ASP) –Reviews - requirements, design, code (starting) –C++ coding rules - automatically checked Objectivity database in use –GEANT 3 data, test beam data, n-tuples –Detector description database ATLAS repository in CVS/SRT –145973 lines of ATLAS FORTRAN –74956 lines of AGE –149099 lines of C++ Documentation on the web - LIGHT Status of ATLAS OO Environment + external code: 332404 FORTRAN 600000 C++ GEANT-4
13
9 November 98 13 Jürgen Knobloch ATLAS Computing ARVE –ATLAS Reconstruction and Visualisation Environment –Developed by Toby Burnett (U. Washington) –Inner Detector pattern recognition - several algorithms in C++ –Muon reconstruction –GUI, Graphics –Generation of single particles --> GEANT4 ATLAS OO projects - Reconstruction
14
9 November 98 14 Jürgen Knobloch ATLAS Computing LHC++: commercial tools - evaluating –Thin HEP-specific layer ATLAS OO projects - Analysis tools
15
9 November 98 15 Jürgen Knobloch ATLAS Computing ATLAS OO Projects - Simulation Simulation: GEANT4 - ATLAS participation and implementation –Test beam simulation under development –Full detector prototype Most US work is done currently within BaBar
16
9 November 98 16 Jürgen Knobloch ATLAS Computing ODBMS (RD45) ATLAS participation –Event storage: GEANT3, GEANT4, test-beam –Detector description database ATLAS OO projects - Data Bases DetectorElement Identifier identify() iterator digits_begin() iterator digits_end() DetectorDescriptor Digit Identifier identify() Point3D position() float response() DetectorPosition Point3D center() Transform3D transform() Basic design of raw data structure:
17
9 November 98 17 Jürgen Knobloch ATLAS Computing Monte Carlo Productions Physics performance work - productions in full swing Many sites moving to Intel platforms - e.g. at CERN&LBL: Bottleneck: People running productions - not infrastructure Next: run reconstruction and produce n-tuples
18
9 November 98 18 Jürgen Knobloch ATLAS Computing Centralized model –all of the event data are stored centrally and coherently managed at CERN –good time response needed Partially decentralized model –replicate the event data at about five regional centers –data transfer via network or movable media CERNRC2CERNRC1 Computing model architecture
19
9 November 98 19 Jürgen Knobloch ATLAS Computing 1 SPECint95 = 40 SPECint92 = 10 CERN units = 40 MIPS CPU requirements Event reconstruction7x10 4 SPECint95 –250 SPECint95-sec/event = 10xFermilab run 1 Monte Carlo production5x10 4 SPECint95 –10% of number of real events Physics analysis15x10 4 SPECint95 –20 analysis groups –25 people in each group => 500 people analyzing –150 simultaneous users –group loops over 10 9 events/month - selecting 1-10% –each physicist loops over selected events 1/month –each physicist loops over AOD/Ntuple 1/day
20
9 November 98 20 Jürgen Knobloch ATLAS Computing Note: The unit cost extrapolation and the requirements both have large errors! The final numbers should therefore be used with great care! Cost
21
9 November 98 21 Jürgen Knobloch ATLAS Computing Definition of milestones Ingredients: Milestones given in CTP OO reconstruction and simulation - first version available end 99 Full chain operational on subset of final infrastructure - 1 year before LHC start-up Coverage of interim needs –Physics studies –Test beam support –Radiation level studies
22
9 November 98 22 Jürgen Knobloch ATLAS Computing Milestones 48 Milestones defined 23 before July 99 Define “near” milestones on a yearly basis Covering tasks in DFDs Tasks for end 99: –read “prototype” or “first version”
23
9 November 98 23 Jürgen Knobloch ATLAS Computing Milestones - 1 General Provide outline of end-99 CTP2/99 Report from review of computing2/99 Install test and verification procedure4/99 Assess development resources4/99 Revisit development environment7/00 Decide on first production OS7/03 Data challenge1/04 Test full chain in real environment7/04 Analysis Analysis requirements collected12/98 Evaluate analysis environment for interim period7/99 Evaluate analysis environment for production period7/02 Implement analysis environment for production period7/03
24
9 November 98 24 Jürgen Knobloch ATLAS Computing Milestones - 2 Event storage Define raw data structure12/98 GEANT3 events in ObjDB8/98 Define initial version of ESD7/99 Define initial version of event tag7/99 Define initial version of AOD7/99 GEANT4 events in ObjDB7/99 Provide DB performance report7/00 Decide on database vendor6/01
25
9 November 98 25 Jürgen Knobloch ATLAS Computing Milestones - 3 Simulation ATLAS simulation domains defined10/98 Define event generator interface1/99 Simulation framework available1/99 Single particles from G4 in ARVE1/99 GEANT4 test beam simulations3/99 Simulation geometry available7/99 First general OO simulation available12/99 Minimum bias overlay5/00 Terminate maintenance DICE3/01
26
9 November 98 26 Jürgen Knobloch ATLAS Computing Milestones - 4 Reconstruction version 1 Reconstruction geometry available3/99 Prepare data12/98 Find tracks2/99 Find ID tracks2/99 Find muon tracks4/99 Find global tracks7/99 Find cal clusters5/99 Calorimeter Calibration4/99 Muon Spectrometer alignment7/99 Inner Detector Alignment7/99 Event display prototype7/99 First complete OO reconstruction12/99
27
9 November 98 27 Jürgen Knobloch ATLAS Computing Milestones - 5 Infrastructure 1 Tbyte database prototype12/98 Conclude study on regional centres11/99 Define financing of computing at CERN10/99 Define scope of regional centres11/99 100 Tbyte database prototype12/03 Full database available12/04 1% processing farm prototype12/02 5% processing farm12/03 40% processing farm12/04 100% processing farm6/05
28
9 November 98 28 Jürgen Knobloch ATLAS Computing Software workshops –1 week, 4 times / year, 1 outside CERN ACOS (Computing steering group) –6 meetings / year (4 during software workshops) DIG (Domain interface group and support team) –8 meetings / year (4 during software workshops) Weekly software meetings –Videoconferenced –Tutorial session about OO issues Working groups: database, graphics, WWCG, … CERN ATLAS Computing Group (ATC) Organisational items
29
9 November 98 29 Jürgen Knobloch ATLAS Computing Conclusions Groundwork has been done, a software process is in place, tools such as GEANT4 have been developed. We are at a turning point from preparatory work to OO software implementation. Physics studies done in FORTRAN have demonstrated the performance of the ATLAS detector. The results will serve as benchmarks for OO software. After physics “TDR” - –move physics studies to OO, get more people on board The distributed (peta-byte) data-mining using regional computing centers needs to be established.
30
9 November 98 30 Jürgen Knobloch ATLAS Computing US-ATLAS contributions (1) US-ATLAS members have played key roles, e.g.: –K. Sliwa >chairing the “Computing Model Group” in preparation of the Computing Technical Proposal; now “World-Wide Computing Group”. >member of ACOS >was the initiator of the MONARC project. –D. Melon >participation in MOOSE project >groundwork for RD45 with “Grand Challenge” –T. Burnett >Major contributor to MOOSE -> developing ARVE
31
9 November 98 31 Jürgen Knobloch ATLAS Computing US-ATLAS contributions (2) US physicists have been involved in the development of the GEANT-4 simulation US laboratories have made major contributions to physics studies and Monte Carlo productions –Computing facilities in several labs –Porting of ATLAS software to specific operating systems I hope that US ATLAS institutes will continue to increase their involvement in ATLAS computing, bringing in their valuable expertise from past and current projects. Early involvement is the key for later success!
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.