Alexander Richards, UCL 1 Atlfast and RTT (plus DCube) Christmas Meeting 18/12/2007

Preview:

Citation preview

Alexander Richards, UCL 1

Atlfast and RTT(plus DCube)

Christmas Meeting 18/12/2007

Alexander Richards, UCL 2

Atlfast

Fast Simulation Package

3Alexander Richards, UCL

ATLAS fast simulation It includes most crucial detector aspects: jets

reconstruction in the calorimeter, momentum/energy smearing for leptons and photons, magnetic fields effects and missing transverse energy

It provides, starting from the list of particles in the event, the list of reconstructed jets, isolated leptons and photons and expected missing transverse energy.

Optionally package provides a list of reconstructed charge tracks

What is Atlfast ?

4Alexander Richards, UCL

Smearing functions replace full chain

4-5 orders of magnitude faster than full simulation

What is Atlfast ?

5Alexander Richards, UCL

How does it perform? Full simulation + reconstruction takes ~½ hr per event. Atlfast test jobs in 12.0.3

104-105 x faster than full chain

2376 s307 sTotal (includes initialisation)

200 ms12.6 msPythia execute per event

21.8 ms8.15 msAtlfast execute per event

ttH(Hbb) (10k, Pythia)

Zee (10k, Pythia)

Sample

6Alexander Richards, UCL

Instructions on web www.hep.ucl.ac.uk/atlas/atlfast (static UCL page) AtlfastDocumentation (Atlas TWiki portal)

Easiest way is to Set up a release directory Set up run time environment (athena) 'get_files XXXXtoAtlfasttoYYYY.py'

XXXX is Pythia or POOL YYYY is CBNT, AOD and in r12 AAN

Configure script 'athena XXXXtoAtlfasttoYYYY.py'

How to run Atlfast?

7Alexander Richards, UCL

Some plots!

High Pt

Low Pt

QCD Jet Sample

8Alexander Richards, UCL

Towards "Atlfast Phase 2"

FastCaloSimParameterised showers in a full calorimeter

FatrasFast tracking via hit simulation

More "Atlas", less "Fast"

9Alexander Richards, UCL

Timing still prohibitive but getting better..... Project is at validation stage

Towards "Atlfast Phase 2"

M. Duehrssen

Alexander Richards, UCL 10

RTT

Run Time Tester

11Alexander Richards, UCL

What is ATLAS RTT?

“The ATLAS RTT is a tool for developers which provides them with a convenient way to test, on a nightly basis, both the status and the output that any changes to their code may have had.”

Does it run? Is the output the same as expected – regression tests/root macros

ATLAS RTT lets you automate:

● Running Athena (as well as non-Athena) jobs.● Running post-job activities i.e. ROOT macros, regression tests and user specified Python scripts.● Publishing all results to a (web served) user specified directory.

12Alexander Richards, UCL

How does it work?

Several NICOS builds occur each night –’Nightlies’ When Complete the build ready flag set and RTT can begin RTT scans all packages in the release for those which require

RTT tests RTT then proceeds in two steps: Firstly the Data Copier checks to make sure that any data

requested by the jobs is available and if not it copies it to the running directory

Secondly the RTT tests are begun and results updated in ‘near real-time’ to the results webpage.

13Alexander Richards, UCL

RTT Results http://atlas-project-rtt-results.web.cern.ch/atlas-project-rtt-

results/weeklyTable/weekly.php

14Alexander Richards, UCL

RTT Results

15Alexander Richards, UCL

RTT Results

16Alexander Richards, UCL

Including RTT jobs

For RTT to pick up that your package requires RTT tests you must add lines to your requirements file.

RTT steered by an .xml configuration file, which must be store it in a subdirectory of your package called ‘test’.

.xml file may be validated on the RTT main page: http://www.hep.ucl.ac.uk/atlas/AtlasTesting/

Every job must belong to a ‘jobgroup’, the name of which must be registered with one of the RTT developers.

Jobs belonging to a job group are treated the same way by the RTT with respect to any post Athena actions/tests.

As many jobgroups as required may be registered to your package.

17Alexander Richards, UCL

Example configuration file

Alexander Richards, UCL 18

DCube

Histogram Comparison

19Alexander Richards, UCL

What is DCube?

DCube is a histogram comparison package. Can be run interactively or as part of RTT Capable of comparing 1D and 2D histos with:

Kolmogorov-Smirnov test test Bin-by-bin comparison

Results generated in user friendly web format

20Alexander Richards, UCL

How does it work?

DCube is steered by .xml configuration file. Config file picked up and run in RTT by a post athena

macro called DCubeRunner .xml file specifies the following:

Absolute location of a reference ROOT file Name of the file to compare with this reference file, called the

Monitored file Which of the three types of tests to perform The P-values to mark ‘warning’ and ‘failure’ threshold for stats

tests Both reference and monitored files may contain multiple

histos each of which may have any of the three tests performed on it

21Alexander Richards, UCL

DCube Results

22Alexander Richards, UCL

DCube Results

23Alexander Richards, UCL

DCube Results

24Alexander Richards, UCL

DCube Results

25Alexander Richards, UCL

Summary

With RTT we have the ability to test the stability of code on a nightly basis.

With the addition of DCube we can now check the ‘stability of our physics’ on a nightly basis

Future plans include creating physics validation plots for both GeneratorsRTT and TestAtlfast packages.

26Alexander Richards, UCL