GlueX Offline Meeting, July 23, 2014

From GlueXWiki
Jump to: navigation, search

GlueX Offline Software Meeting
Wednesday, July 23, 2014
1:30 pm EDT
JLab: CEBAF Center F326/327

Agenda

  1. Announcements
  2. Review of minutes from July 9 (all)
  3. MD5 check of resource files (David)
  4. Haswell CPU testing (David)
  5. Data Challenge 3

Communication Information

Remote Connection

Slides

Talks can be deposited in the directory /group/halld/www/halldweb/html/talks/2014-3Q on the JLab CUE. This directory is accessible from the web at https://halldweb.jlab.org/talks/2014-3Q/ .

Minutes

Present:

  • CMU: Paul Mattione, Curtis Meyer
  • FSU: Aristeidis Tsaris
  • IU: Matt Shepherd
  • JLab: Mark Ito (chair), David Lawrence, Nathan Sparks, Mike Staib, Simon Taylor, Beni Zihlmann
  • NU: Sean Dobbs
  • UConn: Richard Jones

Review of minutes from July 9

We reviewed the minutes.

  • Automatic Tests -- Both the single-track and b1pi tests are now succeeding. Simon checked in a fix to tracking and David a fix to tracking event display.
  • EVIO Build -- Mark has a makefile in scripts/build_scripts that will build the most recent version of EVIO. EVIO has not yet been incorporated into the global build system (i. e., gluex_install).
  • BCAL Timing -- David reported that Mark Dalton has noticed that there is an asymmetry in how the timing is done in simulation and real data for the BCAL. In the simulation the pulse shape is simulated, and a threshold applied to determine the time. In data, a digital constant-fraction discriminator algorithm is used. In the reconstruction a time-walk correction is applied. In the case of the real-data, this correction is largely unnecessary and in fact hurts the time resolution. The nominal plan being discussed in the BCAL group is to emulate the real-data algorithm in the simulation and drop the time-walk correction in the reconstruction.
  • Global Event Time Smearing -- Richard has started work on this on a branch of the Subversion repository. Paul has looked at incorporating the effect in his analysis library, but thought that the start counter time resolution should be more realistic that it is now to make progress. Others thought that even a constant RMS time smearing for all positions in the counter would be sufficient to make progress on the problem of picking out the right RF bucket. Turns out there is a command line parameter for mcsmear to set the this smearing constant, "START_SIGMA".
  • Truth vs. Hits -- Richard reminded us that the parallel scheme of truth information and smeared hit information is already in place for the CDC, FDC, FCAL, and BCAL. He is making the scheme global, adding it to the start counter for example.

MD5 Check-Sums and Resource Files

David described the problem that Matt ran into where the downloading of resource files (in his case the magnetic field map) was interrupted, resulting in a corrupt file that crashed programs that used it. The solution is to use the MD5 check-sum that is generated and downloaded along with each resource file to check file integrity each time the file is read. David is working on implementation for the next version of JANA.

Haswell CPU Testing

David reported on tests he has done bench-marking a demo machine that SciComp had on loan for this purpose. He compared the demo Haswell machine with an Ivy Bridge gluon machine in the Counting House.

Parameter Haswell Ivy Bridge
CPU Chips 2 1
cores/chip 18 16
hyper-threads per core 2 2

He did two types of comparisons, one with a dummy, but CPU-intensive, JANA application and another with bggen data, looking at hdgeant, mcsmear, and hd_ana, one by one. See his slides for the plots. He sees big improvements with the new Haswell architecture, as much as 71% per core (for mcsmear). Richard remarked that this might be more improvement that we have a right to expect based on a quick Google survey. There were several questions about the exact configuration of the new system (and of the old for that matter) that might affect the comparison, but the results are nonetheless encouraging and may save "us" some money.

Data Challenge 3

David reported that the complete chain from simulated data in HDDM format to EVIO format to DANA classes is ready to go. The current system only creates EVIO data in single-event form (as opposed to multi-event block form) [terminology?]. Mark will start to test this part of the chain, i. e., creating simulated raw data and putting it on tape. He will also start working on a new software stack to do reconstruction starting from the tape files.

Build Items

  • Paul has prompted Mark to take up work again on a global build system for GlueX. Paul would like to see multiple versions of each package co-exist in the tree with dependencies among them taken care of.
  • Richard seconded Matt's suggestion that we delete makefiles from the sim-recon. Mark volunteered to do so.
  • Richard requested that someone convert the build of hdds from GNUMake to scons. David agreed to look into what is involved.