Analysis Queues Performance

America/Chicago
Vidyo

Vidyo

Ilija Vukotic (Universite de Paris-Sud 11 (FR))
Description
There is a need to understand performance of ATLAS analysis queues, and eventually improve it. We will start with US sites and try to document our findings. Hopefully our experience will help to all ATLAS grid sites. The two things we will try to steer clear off: 1. Micromanaging sites. 2. Micromanaging ATLAS code. We'll use continuously running HC functional test jobs to look at performance of each site, try to understand any features we observe. While direct comparison between sites is not possible, we'll try to bring them all to it's optimal performance.
present: shawn, david, ilija, sarah, wei
  1. aglt2 24 core mystery.    - only my HC jobs affected. only after switch to  direct access. job times-out. look again at log file. try to manually set to copy to see if that will help.
  2.  HU. still bad performance  - will need ont-on-one discussion
  3. BNL - no news. - issue was a long pre-stage time.
  4. MWT2 - problem with the jobs not being submitted has been solved
  5. Patrick's request to have ANALY_SWT2_CPB and ANALY_OU_OCHEP_SWT2 in the tests has been accomodated.
  6. did not start yet switching direct access/pre-stage tests
  7. having plots of currently running and currently queued jobs: there is info at panda (http://panda.cern.ch/server/pandamon/query?tp=queue&id=ANALY_MWT2) but this is set manually by each site. Nobody knows which of the filelds are used by panda. Need to contact panda people to get access to real-time info.
  8. simple and easy to understand weekly plots for sys admins to look at are mostly there but still not finished.
  9. automatic e-mails - nothing done yet
  10. dcache billing db: access to log files- ct2-dc4.uchicago.edu:/opt/d-cache/billing/ . Will send me a link to extracted info. Will get read only access to the AGLT2 billing db from Shawn. Some queries might not work.
  11. we moved to compiled read script. Have to handle 0's in a beter way.

from Sarah email:
As we discussed, here is data extracted from the billing logs at
MWT2.  The fields are date, time, pfn, file size, bytes read, and
milliseconds for the transfer.  The data is for Apr 1 - Apr 12.
http://www.mwt2.org/~sarah/billing/data.log.tar.gz
If you browse this directory, you also see the daily data.
http://www.mwt2.org/~sarah/billing/2012/04/



There are minutes attached to this event. Show them.
    • 14:00 14:10
      review of open issues, current performance 10m
      Speaker: Ilija Vukotic (Universite de Paris-Sud 11 (FR))
    • 14:30 15:00
      AOB