ATLAS Physics Analysis Framework James R. Catmore Lancaster University

Introduction  Physics analysis is the final link in the ATLAS chain  It is the point at which data is “handed over” from technical experts to the physics community  As such it sets significant challenges on the software infrastructure  Getting the data to the physicists across the world  Packaging it in a readily readable form  Providing tools which enable users to perform the sophisticated studies which will be necessary to discover new physics  Providing computing resources to enable large-scale data processing  In this talk:  Overview of analysis framework components  Why should we be using the Grid to do physics analysis?

Overview of the analysis framework (steady state) Monte Carlo Detector Reconstruction Digits AOD AOD Building ESD TIER 1 SITES ESD AOD TIER 2 SITESGRID MIDDLEWAREUSERS Analysis jobs Results TIER 0 (CERN) / PRODUCTION SYSTEM

Terms and tools  ESD: Event Summary Data  Will not in general be used for analysis  AOD: Analysis Object Data  This is the data which will be used by physicists for most analyses  Persistency: format in which the data is written to disk  ATLAS uses a format known as POOL  Athena: the overarching software framework within which all tasks are run  Analysis code is implemented as Athena algorithms  Analysis tools: software within Athena which expedite efficient physics analysis  “The Grid”: distributed network of storage and computing facilities where ATLAS data will be stored and on which analysis will be performed  Also includes “middleware” which provides the interface between the user and the grid sites  Jobs: Athena tasks set by the users to be performed on the Grid  Job submission tools: software which facilitates the submission of jobs to the Grid  ATLAS and LHCb share a common tool - GANGA  The ATLAS tool for managing Grid files is called DQ2 (Don Quichote)

Analysis Object Data (AOD)  Distilled information from the event reconstruction  Makes strong use of inheritance  Ultimately inherit from a four-momentum implementation  AOD objects are designed to reflect the physical objects which they represent  Muon, electron, photon, track, missing Et, tau-jet, b-jet  Common “look and feel” to all classes  Data is grouped into events and packaged in STL containers  MuonContainer, ElectronContainer, TrackParticleContainer etc  The basic operation in the analysis code is therefore looping over the AOD objects in these containers and interrogating them for information

Analysis Tools and event selection  Software which facilitates the writing of clean and transparent analysis code  Sorting, selecting, filtering, combining, calculating common physical quantities, removing overlaps  Principal tools  AOD analysis tools  EventView  B-physics analysis package  Event selection  Data will not come out of ATLAS nicely packaged according to the signal event type  Users will need to select events according to simple criteria  E.g. “event contains two muons with pt > 6GeV”  This information is known as “metadata” is implemented as a “TAG” on each event  Accessed through AMI, the Atlas Metadata Interface

“Post analysis” tools

Using the Grid to do physics analysis

Why use the Grid for analysis?  “Why can’t I just copy the files to my University and run analysis locally?”  Well, at the moment, with conveniently packaged sets of Monte Carlo, that’s possible  Once ATLAS starts to produce data, there will simply be too much of it for local analysis  Storage space  Time for copying files across to local institutes  So whilst it may not be immediately necessary to use the Grid, it makes sense to learn how to use it in advance of it becoming essential  The Grid provides immense computing resources which enable a user to run hundreds of jobs simultaneously  Users do not have to worry about installing software at their institutes  The DQ2 database provides an easy method of locating the required data  Users don’t have to worry about where it is  GANGA provides a very simple interface to the Grid; anyone who can run Athena can use the Grid  Graphical user interface provided

What do I need to use the Grid 1.A Grid Certificate 2.Membership of the ATLAS Virtual Organisation 3.A computer set up as a Grid User Interface (UI) Comprehensive workbook instructions (S. Lloyd):

About GANGA  Gaudi, Athena and Grid Alliance  Joint ATLAS/LHCb grid job submission tool  Minimizes user contact with Grid fabric  Principally designed for analysis but also for small-scale private Monte Carlo  Automatically retrieves and registers files  Can operate on either the Grid or a local batch system  Provides a python command line or a GUI  Easily installed on local machines Main page and installation instructions: Most recent tutorial (September 2006):

A complete example Set up Grid UI and DQ2: Search for the dataset:

DQ2 web interface

A complete example (ii) Set up GANGA and CMT: Check out analysis package: Setup in cmt directory of package:

A complete example (iii) Start GANGA:

A complete example (iv) Define the job:

A complete example (v)

A complete example (vi) Get the n-tuple: …which drops the results into the /afs home directory. These can then be analyzed in ROOT as normal

Other facilities and current issues  Job splitting  TAG analysis  Local backend switch  User-defined Monte Carlo production  Seamless registration onto DQ2  Ability to read old LFC data  GUI  Issues  DQ2 can only copy with datasets at one site  Not a GANGA issue but affects the way we need to work at the moment  Request to have the facility to view jobs as they run

Ganga GUI

Conclusions  The Physics Analysis Framework for ATLAS is now in an advanced state and will be ready for data taking  Physicists will need to use the Grid to do their analysis  Most of the tools are ready to be used - it makes sense to learn how to use them now  A large quantity of high-quality documentation is available  A tutorial will be held in the UK within a few months - announcements will be made shortly