Implementing Metadata Using RLS/LCG

1
Implementing Metadata Using RLS/LCG

• James Cunha Werner
• University of Manchester
• http//www.hep.man.ac.uk/u/jamwer/

2
Babar Experiment

• The BaBar experiment studies the differences
  between matter and antimatter, to throw light on
  the problem, posed by Sakharov, of how the
  matter-antimatter symmetric Big Bang can have
  given rise to todays matter-dominated universe.
• High energy collisions between electrons and
  positrons produce other elementary particles,
  giving tracks and clusters which are recorded by
  several high granularity detectors and from which
  the properties of the short-lived particles can
  be deduced.

3

• Each recorded collision, called an event,
  comprises a large volume of data, and thousand of
  millions of events are recorded, giving a total
  dataset size of hundreds of thousands of
  Gigabytes (or hundreds of Terabytes).

4
Sources of Data in Babar
5
Amount of data
Files Size (TB) Events (Million)
Run1 6,972 2.0 593
Run2 11,527 6.3 1,925
Run3 7,383 3.2 951
Run4 16,671 12.2 3,999
Run5 (2xRun4) ??? 32,000 24 8
Run6 (2xRun5) ??? 64,000 48 16
Run7 (2xRun6) ??? 128,000 100 32
SuperBabar !
Systematic errors gtgtgt statistical errors
Same amount of Monte Carlo Generated data!
6
Data Structure

• The user interface to the eventstore event
  "collection". Each collection represents an
  ordered series of N events and a user can choose
  to read the events from the 1st one in the
  sequence or from any given offset into the
  sequence.
• Data components
• hdr - event header
• usr - user data
• tag - tag information
• cnd - candidate information
• aod - "analysis object data"
• tru - MC truth data (only in MC data)
• esd - "event summary data"
• sim - "sim" data from BgsApp or MooseApp like
  GHits/GVertices (only in MC data)
• raw - subset of raw data from xtc persisted in
  the Kanga eventstore

7
Data organisation

• How data are stored (level of detail)
• micro hdr usr tag cnd aod ( tru)
• mini micro esd

• Data access
• collections - these are "logical" names that
  users use to configure their jobs. These are
  site-independent so (assuming the site has
  imported the data) the same collection name
  should work at any site.
• logical file names (LFN) - these are
  site-independent names give to all files in the
  eventstore. Any references within the event data
  itself _must_ use LFN's so that these remain
  valid when they are moved from site to site.
• physical file names (PFN) - these are file names
  that will vary from site to site. In practice
  they are usually derived from the LFN's by adding
  a prefix that encapsulates how the data is
  accessed at that site.

8
(No Transcript)
9
Feeding RLS with metadata

• Generation of basic metadata file with files
  selection!/bin/bashBbkDatasetTcl
  --dbsitelocal gt MetaLista.txtcat MetaLista.txt
  awk '// print "BbkDatasetTcl --site local
  --nolocal \""1"\""' gtgt geratclchmod 700
  geratcl./geratcl
• Feeding RLS with basic files
• !/bin/bashls .tcl awk '// split(1,a,".")
  print "edg-rm --vo babar cr file///home/jamwer/Pg
  mCM2/MetaData/"1 " -l lfn"a1 " gt "
  a1".rlstok"' gtgt alimrlschmod 700
  alimrls./alimrls

10
Conformity CE catalogue

• Run evaluation software to establish CE
  conformity and perform catalogue update.
• !/bin/bashldapsearch -x -H
  ldap//lcgbdii02.gridpp.rl.ac.uk2170 -b
  'Mds-vo-namelocal,oGrid' '((objectClassGlueCE)
  (GlueCEAccessControlBaseRuleVObabar))' grep
  "GlueCEUniqueID" gt cenames.txtcat cenames.txt
  awk '// print "./catal "2' gt subload.shchmod
  700 subload.sh./subload.shcat loadrlssubm gtgt
  1.histocat 1.histo awk ' /Sub/ FileName2
  /https/ HandleName2 print "echo " HandleName
  "gt " FileName".tok " ' gtgt gridtokchmod 700
  gridtok./gridtok

11
Conformity validation

• Verify if site follow experiment standards
• !/bin/bashecho Hostname
  /bin/hostnameecho Start time
  /bin/dateecholocalpwdecho Babar
  initialisation ". VO_BABAR_SW_DIR/babar-grid-set
  up-env.shechoecho Environment
  variables"printenvechocd localecho Arquivos
  disponiveis locallsechoecho " - - - - - -
  - - - - - - - - - - - - - - - - - - - - - - - - -
  - - - - - - - - - - - - - - - - "echocd
  BFDIST/releases/14.5.2srtpath 14.5.2
  Linux24RH72_i386_gcc2953cd localBbkDatasetTcl
  --dbsitelocal gt MetaLista.txtcat MetaLista.txt
  awk '// print "BbkDatasetTcl --site local
  \""1"\""' gtgt geratclchmod 700
  geratcl./geratclexport CE_NAME1ls .tcl
  awk -v siteCE_NAME '// split(1,a,".") print
  "edg-rm --vo babar addAlias cat " 1"
  lfn"a1"."site ' gtgt alimrlschmod 700
  alimrls./alimrlsechoecho " - - - - - - - -
  - - - - - - - - - - - - - - - - - - - - - - - - -
  - - - - - - - - - - - - - "echoecho End time
  /bin/date

12
Analysis Submission to Grid
(Prototype)

• Single command ./easygrid dataset_name
• Perform Handlers management and submission
• Configurable to achieve users requirements
• Software based in State-machine
• Verify skimdata available
• If not available perform BbkDatasetTCL to
  generate skimData. Each file will be a job.
• Verify if there are handlers pending
• If not, script generation (gera.c) with
  edg-job-submit and ClassAdds, and script
  execution. Nest for submission policy and
  optimisation.
• If yes, verify job status. When the all jobs
  ended, recover results in user folder.

13
Job Submission system, metadata and data
14
Metadata/Event files and Computer elements
For each dataset there is a metadata file
containing the names of the event files. These
physical files are registered with the RLS, with
several logical file names in the format
datsetname_CEJobQueue assigned to them as
aliases, showing the CEs which contain copies of
that dataset. Searching all the aliases for a
dataset name provides a list of CEs to which jobs
can be submitted.
15
Managing large files in Grid

• The analysis executable is allocated in the SE
  and its logical file name (LFN) is also
  catalogued in the RLS so any WN need download it
  only once.
• Metadata not only for data, but to support other
  files as well.

16
Gera

• Generation of all necessary information to submit
  the jobs on the Grid.
• Job Description Language (JDL) files
• the script with all necessary tasks to run the
  analysis remotely at a WN
• some grid dependent analysis parameters.
• The JDL files define the input sandbox with all
  necessary files to be transferred
• WN balance load algorithm matches requirements to
  perform the task optimally.

17
Running analysis programs
When the task is delivered in the WN, scripts
start running to initialize the specific Babar
environment, and the analysis software is
downloaded.
18
Benchmarks
Behavior of particles in the BaBar
Electromagnetic Calorimeter (EMC)

• The different behavior of electrons, hadrons, and
  muons can be distinguished.
• Performing this analysis takes 7 days using one
  computer 24 hours a day.
• Using 10 CPUs in parallel, accessed via the Grid,
  it took only 8 hours.

19

• Pi- N Pi0 decays, with N 1, 2, 3 and 4

• Invariant masses of pairs of gammas, as measured
  by the EMC, from Pi0 decay produce a mass peak at
  135 MeV (the peak in the plot). All other
  combinations are spread randomly around all
  energies (background).
• There were 81,700,000 events in the dataset and
  it took 4 days to run in production, with 26 jobs
  in parallel to run it in one single computer
  would take more than 3 months.

20
Summary

• Easygrid is working and provides all job
  submission structure using LCG grid, RLS and
  metadata management.
• Provides handlers management transparent to the
  user.
• Easy to use !!!
• Configurable to achieve users requirements and
  maybe for other experiments as well.
• See homepage http//www.hep.man.ac.uk/u/jamwer/
  for more details.
• Thanks for the opportunity!