GlueX Offline Meeting, March 23, 2011

From GlueXWiki
Jump to: navigation, search

GlueX Offline Software Meeting
Wednesday, March 23, 2011
1:30 pm EDT
JLab: CEBAF Center F326/7

Agenda

  1. Announcements
  2. Review of minutes from the last meeting: all
  3. SIMD
    • speed enhancement (PDF) David L.
  4. Floating point entropy: Richard
  5. Make system, includes, in JANA and sim-recon: Matt
  6. BCAL studies and build issues: Andrei and Irina
  7. FFREAD modifications: Sascha
  8. Action Item Review: all
    1. Resolved issues
    2. Outstanding issues
    • To view these links, login first. Anonymous login is available.
  9. Review of recent repository activity: all

Communication Information

Telephone

To connect by telephone:

  1. dial:
  2. enter access code followed by the # sign: 3421244#

Video Conferencing

Slides

Talks can be deposited in the directory /group/halld/www/halldweb/html/talks/2011-1Q on the JLab CUE. This directory is accessible from the web at https://halldweb.jlab.org/talks/2011-1Q/ . You have to be a member of the "halld" Unix group to do this.

Minutes

Present:

  • CMU: Will Levine, Curtis Meyer
  • IU: , Mihajlo Kornicer, Kei Moriya, Matt Shepherd
  • JLab: Mark Ito (chair), David Lawrence, Simon Taylor, Elliott Wolin
  • UR: Zisis Papandreou, Andrei Semenov

(not sure I got everyone...let me know if your name is missing)

Announcements

  • Elliott announced that there are two CNU students starting on the Java-based event display.
  • Matt is developing a new cluster finder for the BCAL. It is work in progress at present.

Review of minutes from the last meeting

We went over the minutes from the March 9 meeting. Zisis recommended that in the re-organization of the Offline Software wiki page, the instructions on how to get started should be easily located and their contents reviewed and updated. Newcomers should know what they should be looking at and the material should be accurate.

SIMD speed enhancement

David showed a histogram of the event analysis rate for several intervals during a reconstruction run. He made the plot for code with SIMD instructions enabled and without. He repeated the exercise for other combinations of number of cores (on the machine) and number of threads with consistent results. He concludes that the SIMD instructions are giving us about 5% to 10% improvement in execution speed.

We agreed that given this situation, leaving the default state of the code with SIMD-off is sensible.

Floating point entropy

Recently Richard diagnosed a problem that was causing segmentation faults in our reconstruction code. Richard made a proposal for a review of our code looking for problematic instances of comparisons of floating point numbers. We decided that we needed Richard to lead this discussion and tabled the issue for now.

Make system, includes, in JANA and sim-recon

Matt re-capped his proposal that the make system use only include files that exist in the source distribution tree and not rely on files from an "installed" location. That way a recent check-out will not have to have any "make install" prerequisites before "make".

We had some discussion. Mark thought that the proposal made sense. David pointed out that if one does from-the-top make, then all the necessary installs will be performed. He also pointed out that the sim-recon code has a slightly different usage pattern, as it is usually the end product build, not really a package to be used by other packages, therefore the "install" could or should be different than from other cases.

We did not come to a conclusion. More discussion is needed.

BCAL studies and build issues

At the March 10 BCAL Readout meeting Andrei and Irina reported difficulty using our offline software and it was suggested that the issues be discussed at this meeting. Since then the problems have been resolved, but there were a couple of lessons learned:

  • Our "getting started" documentation needs to be updated (see item above on wiki page re-organization).
  • Mark's build scripts need better documentation.

FFREAD modifications

We briefly discussed issues surrounding the use of FFREAD cards in our GEANT3-based simulation, in particular an issue with decimal points and one on the number of characters in a card identifier. See the thread rooted on this message in the email archive. We decided for now there was no action to be taken; it is best that we just learn to use the code as it is.

New Action Items

  1. Decide on Matt's include-related proposal for the make system(s).
  2. Our "getting started" documentation needs to be updated.
  3. Mark's build scripts need better documentation.