CMS Software

Vincenzo Innocente: CMS Software

Introduction for Summer Students, 13th July 2004

CMS Software

Physics Analysis in a Brave New Woorld

Vincenzo Innocente(original form Stephan Wynhoff)




OverviewNot a talk to describe why things are the way they are

1. The main projects

and tools

2. The role of the central framework, the persistent store

3. A glance on how to use it

COBRAIGUANA

ORCA

FAMOS

LCG/AA

OSCAROVAL

SCRAM

IGUANACMS

IGNOMINY



Challenges: Geographical Spread

1700 Physicists 150 Institutes 32 Countries(and growing!)

CERN state 55 %

NMS 45 %

Major challenges associated with:Communication and collaboration at a distanceDistributed computing resources Remote software development and physics analysis



Tier 1

Tier2 Center

Online System

Offline Farm,CERN Computer Ctr

> 20 TIPS

FNAL CenterIN2P3 Center INFN Center RAL Center

InstituteInstituteInstituteInstitute ~0.25TIPS

Workstations

~100 MBytes/sec

~2.5 Gbits/sec

100 - 1000 Mbits/sec

Physics data cache

~Pbyte/sec

~2.5 Gbits/sec

Tier2 CenterTier2 CenterTier2 Center

~622 Mbits/sec

Tier 0 +1

Tier 3

Tier 4

Tier2 Center Tier 2

Experiment

Software to● keep the systems running● distribute data & jobs ● simulation, reconstruction, analysis

CMS Computing Model



(Some of the)Major Challenges for LHC SW

● Events are big (raw event is 1MB, with Monte-Carlo 2MB)● Detector digitization has to take into account multiple crossings

● Pileup: @1034 = 17 minimum bias events/crossing● Calorimetry needs -5 to +3 crossings● Muon DT ought to have +- 10 crossings● Tracker loopers can persist for many crossings

● Typically need info from ~ 200 mb events per signal event● Study at different luminosities infers different pileup

● include pileup in digitization (front end of reconstruction)● Track finding in very complex environment● High magnetic field and ~ 1 rad length of tracker material:

● Lots of bremsstrahlung for the electrons, ● TK-ECAL matching non-trivial



HZZ ee event with MH= 300 GeV

1032 cm-2s-1 1033 cm-2s-1

1034 cm-2s-1 1035 cm-2s-1



Pileup● For 1 million Signal events need 200 million min bias events

● Impossible with current CPU, storage etc● Solution is to sample a finite number pseudo-randomly● Problems can come when one single mb event by itself would

trigger the detector● You would get this trigger many many times● Filter the minimum bias events, but remember to take into account the

removed events● must sample from full range of mb events to ensure patterns do not

repeat too often● 200mb events = 70MB

● massive data movement problem● Pileup is CPU intensive



Simulation, Reconstruction, Analysis

Monte-Carlo generator

Ntuple

ZebraFZ

CMSIM

CMSIM

CMSjet

User Analysis

Monte-Carlo generator

OODB

OSCAR

ORCA

User Analysis

FAMOS

FAMOS



OSCAR

HEPEVTNtuple

ORCA

OODBminbias

OODBsignal

OODBDigis

SimReader RecReader

MC generator

CMKIN

Production

Production User

The Analysis Chain

Generation Simulation

Digitization Reconstruction

OODBSimHits

Analysis

OODBRecObs (DST)

Histo



Simulation

● Understand detector response

● optimise geometry● set-up, test alignment● understand signal and

background topologies● feasibility studies of

analyses● optimise trigger● optimise physics

selections

● Mainly FORTRAN (some non-standard)

● use GEANT3, Zebra

● (almost) out of use

● OO design

● use GEANT4

● integrated with CMS software (COBRA)

Tasks CMSIM

OSCAR



OSCAR in 2004

Official Simulation Production Program

● Create and write to OO database and read with ORCA

● User Interface (.orcarc) compatible with ORCA

● Performance:

● Cuts per volume with same values as in CMSIM

● Magnetic field tracking tuned

Two times slower than CMSIM (133)

Released version 3.3.0 Capabale of simulating several million events.



OSCAR Information

● Albert De Roeck et al.

● http://cmsdoc.cern.ch/oscar

● [email protected], [email protected]

● Meetings: SPROM = Simulation PROject Managementevery 2 weeks Monday 16:30h in 40-2A-01



Detector Description DataBase



DDD Information

● Michael Case

● http://cmsdoc.cern.ch/cms/software/ddd/www/index.html

● [email protected]



Geometry

● Collects XML description of the CMS detector

● Materials● Positions● (cuts for OSCAR)

● Contact: Pedro Arce

● http://cmsdoc.cern.ch/cms/software/geometry/index.html

CMSIM 133 Geometry 182



Reconstruction

● Combination of Signal & Pile-up events

● Detector digitisation

● (this is the stuff that comes out of the detector)

● Reconstruction of detector and simple objects

● Tracks, Clusters, Vertices

● Reconstruction of physics objects

● Jets, Electrons, Photons, Muons

● Simulation of L1 Trigger decisions

● The Higher Level Trigger algorithms



Digitisation and Pileup

● High luminosity -> 17 minimumbias events in one bunch crossing

● Overlay crossings -5 to +3

● 200 min.bias for 1 signal event

● "recycle" min.bias events



Stages of Reconstruction

SimHitsProduced by Geant, stored in DB

Digis Include Pileup, some stored in DB (Tk)

RecHitsPre-processed digits,some stored in DB (Calo)

RecObjTracks,Clusters etc,some stored in DB

4vectors Produced by MCstored in Ntuple



ORCA● the flagship OO software project

● started in Sept. 1998

● currently in eighth major release (8.2.0)

● widely used by physicists ● HLT studies● DAQ TDR● Physics TDR● DC04

● not everything can be stored in the database ● Storing with POOL root-streaming

● a prototype is not a final product



ORCA ApplicationsFully embedded in COBRA framework, currently:

1. G3Reader (Also called ooHits)1. Read cmsim fz files

1. Write cmsim hits to Db

2. SimReader (also called ooDigi)1. Pileup arbitrary numbers of crossings pseudo-randomly2. Full digitization

1. persistent storage of results

3. RecReader_ Read and write Digis and RecObj (DST)1. Calorimetric clustering 2. Jet finding (with any types of objects)3. Muon Segment and track finding4. Track finding5. Primary vertices



ORCA Project Structure



ORCA Information

● Stephan Wynhoff et al.

● http://cmsdoc.cern.ch/orca

● [email protected], [email protected], [email protected]

● Meetings: RPROM = Reconstruction PROject Managementevery 2 weeks Monday 16:30h in 40-2A-01



COBRA

CMS Analysis and Reconstruction Framework

● Glue it all together

● Insulate user code from services

● Manage persistent data transparently

● user code does not see any ROOT/IO related code

● Manage Collections, Runs etc

● Manage the order of reconstruction

● Ensure a uniform interface to all CMS code



Federation

wizards

Detector/EventDisplay

Data Browser

Analysis jobwizards

Generic analysis Tools

ORCAORCA

FAMOSFAMOS

LCGLCGtoolstools

GRIDGRID

OSCAROSCARCOBRACOBRA

DistributedData Store

& ComputingInfrastructure

CMSCMStoolstools

Architecture Overview

ConsistentUser Interface

Coherent set of basic tools and mechanisms

Software development and installation



COBRA Components

ODBMS(POOL) Geant3/4

SEALCLHEP

ROOTC++

standard libraryExtension toolkit

Reconstruction Algorithms

Data Monitoring

Event Filter

Physics Analysis

CalibrationObjects Event Objects

Generic Application

Framework

Physics modules

Utility Toolkit

SpecificFramework

CMS adapters and extensions

ConfigurationObjects

GeometryObjects

GRID tools



COBRA Information

● Vincenzo Innocente et al.

● http://cern.ch/cobra




IGUANA● browse federations

● display stored histograms

● event display



OSCAR Visualization• Interactive Geant4 3D CMS detector geometry: physical volume

tree;• Interactive overlap detection: find overlaps and show result details

in list.



ORCA Visualization• Interactive 3D CMS detector geometry (Geant3) for sensitive volumes

with levels of details;• Interactive 3D reps of reconstructed and simulated events including

visualisation of physics quantities such as tangent of a simhit;• Access event by event or automatically fetching events (no batch

mode);• Event and run number displayed.

• Interactive 3D CMS detector geometry (Geant3) for sensitive volumes with levels of details;

• Interactive 3D reps of reconstructed and simulated events including visualisation of physics quantities such as tangent of a simhit;

• Access event by event or automatically fetching events (no batch mode);

• Event and run number displayed.

• Multiple (cloned) views;• Slices and cuts;• Printout for selected object;

• Zoom and search;

• Context help;• Viewpoints.



IGUANA Information

● Lucas Taylor, Ianna Osborne, Lassi Tuura

● http://cern.ch/iguana




Fast MonteCarlo Simulation

● FAMOS for OSCAR

● less geometry volumes

● less detailed GEANT4 options

● FAMOS for ORCA

● faster algorithms

● FAMOS for ORCA

● simple parametrisation of resolutions & efficiencies

● tuned to full simulation/reconstruction

Working for all sub-detectors, soon also for HTL

MC 4-vector

SimHit

RecHit / Digi

Analysis obj

100-200 sec

1-10 sec

10-100 sec



FAMOS Information

● Patrick Janot

● http://cmsdoc.cern.ch/famos




Project Relations

● All Projects listed onhttp://cmsdoc.cern.ch/cmsoo/cmsoo.html

COBRA7.8.5

IGUANA5.0.1

ORCA8.2.0

FAMOS0.9.0

LCG/AA

OSCAR3.3.0

IGUANACMS1.9.0

IGNOMINY1.9.0



External Projects



SCRAM● manage our code building and configuration

(more reproducable than make and autoconf)● typical commands are

● scram listThis gets the list of all the current code releases known to scram

● scram project ORCA ORCA_8_2_0This gets scram to create a local area for you that contains all the directories and configuration files you will need in order to work with this release of ORCA

● eval `scram runtime -csh`set required environment variables

● scram bThis gets scram to compile and/or link the code

● Information● Shaun Ashby● http://cmsdoc.cern.ch/Releases/SCRAM/current/cgi/scrampage.cgi● [email protected]



Varia

● CVS

● that is where the code is stored

● maintains many versions of one file

● full history, parallel development

● OVAL

● tool to automatize validation

● Ecole Polytechnique

● http://polywww.in2p3.fr/cms/software/oval/

● Savannah

● Feature (bug) reporting system

● http://savannah.cern.ch



Pure C++● We work entirely in C++

● You can use any standard tools of C++

● COBRA pins DB objects in memory for duration of event

● You access C++ objects

● The fact that they may have be transient or in the DB is invisible to

you

● This is true for most “developer” code and all “user” code

● Direct ROOT usage only for very private code● Avoid it if you think your code will ever become official

● You can change local copies of the data, but you can’t

unconsciously change persistent data even as viewed by another

module in this reconstruction job.



Generic WWW

● Main CMS page http://cmsdoc.cern.ch/cms.html● Public relations http://cmsdoc.cern.ch

● OO software http://cmsdoc.cern.ch/cmsoo/cmsoo.html

● Finding MC events (very little for ORCA-7) http://cmsdoc.cern.ch/cms/production/www/html/general/index.html

● PRS groups http://cmsdoc.cern.ch/cms/PRS/www/prs.php● meetings every 2nd Tuesday/Wednesday



CMS Software● There is much that will be very new● But you should be able to quickly do the things you were used to do● And the future possibilities are exciting and can offer you and CMS a

significant advantage to get at the Physics first.● C++/OO code can be hard to understand

●Not everything works as intended! ●Be a little patient also, sometimes the thing that seems to you the highest requirement may have not yet reached the top of the stack.●All meetings RPROM, SPROM, PRS are open to all and video-conferenced at times that, while not always convenient, are not impossible for most time-zones

●Everyone has a steep learning curve to followEveryone has a steep learning curve to follow●Use the documentation toolsUse the documentation tools●Ask people, they want to helpAsk people, they want to help



Summary● LHC is extremly demanding on software● Object Oriented techniques will help in answering that

challenge● CMS is well advanced deploying Tools for Simulation,

Reconstruction and Analysis● Physicists can work successfully with the Software● Summer Students can contribute!

New Woorld for the Brave

Brave New Woorld

CMS Software

Documents

Transcript of CMS Software