[Halld-offline] Offline Software Meeting Minutes, August 31, 2016

[Mark Ito]

Ladies and Gentlemen,

Please find the minutes below and at 
https://halldweb.jlab.org/wiki/index.php/GlueX_Offline_Meeting,_August_31,_2016#Minutes 
.

   -- Mark

____________________________________


    GlueX Offline Meeting, August 31, 2016, Minutes

Present:

  * *CMU*: Naomi Jarvis, Curtis Meyer
  * *IU*: Matt Shepherd
  * *JLab*: Alexander Austregesilo, Brad Cannon, Mark Ito (chair), Paul
    Mattione, Nathan Sparks, Simon Taylor, Beni Zihlmann
  * *NU*: Sean Dobbs
  * *UConn*: Richard Jones


      Announcements

 1. sim-recon release: version 2.4.0
    <https://mailman.jlab.org/pipermail/halld-offline/2016-August/002439.html>.
    Mark reminded us that this came out August 24. It is exactly the
    same as offmon-2016_02-ver06.
 2. build_scripts release: version 1.2
    <https://mailman.jlab.org/pipermail/halld-offline/2016-August/002443.html>.
    Mark notes in his email that the idea is make it easier for folks
    working at JLab to set-up an environment that is the same as some
    "standard" version, except that a custom, user-defined build of
    sim-recon is used.


      Review of minutes from the last meeting

We went over the minutes from the meeting on August 17 
<https://halldweb.jlab.org/wiki/index.php/GlueX_Offline_Meeting,_August_17,_2016#Minutes>. 


  * Again, no new work on bending the curve on plugin growth in
    sim-recon. This does seem to be a real issue; it affects build times
    and disk footprint. Nathan noted that for people who are building
    private version of sim-recon and are trying to save on disk space,
    one can switch off inclusion of debugging symbols on the SCons
    command line:

scons DEBUG=0

  * sim 1.1 is completely done. Those last 20 jobs completed successfully.
  * Mark spoke with David about taking advantage the new HDDM
    multi-threaded I/O under JANA. It would indeed take some
    non-negligible work. We don't have a volunteer to do this at
    present, but the demand does not seem high at this point either.


      Report from the August 25th SciComp Meeting

Mark gave the report. Of the items on the agenda 
<https://halldweb.jlab.org/talks/2016/SciComp_2016-08-25.pdf>, most of 
the time was taken with the first item. Chip Watson has circulated a 
white paper 
<https://halldweb.jlab.org/talks/2016/Scientific%20Computing%20Allocations%20Process.pdf> 
describing a proposal for peer-review-based planning and allocation of 
computing resources on the JLab farm. Mark encouraged us to look over 
the proposal as a basis for future discussion and feedback. Major 
changes to the way we do planning for farm use and how that use is 
managed could start appearing soon.

[Added in press] Graham Heyes has also circulated an email initiating 
discussion 
<https://halldweb.jlab.org/talks/2016/computing_resource_scheduling_Heyes.pdf> 
within Experimental Nuclear Physics at JLab.


      Launches


        Last Monitoring Launch

Alex reviewed his announcement of completion of the monitoring launch 
<https://mailman.jlab.org/pipermail/halld-physics/2016-August/000742.html>. 
This launch tested two new capabilities: use of ROOT 6 and the new EVIO 
parser from David Lawrence.

Independently, Alex did a comparison of the launch code compiled against 
ROOT 5 versus the same code compiled with ROOT 6. The resulting 
histograms were identical.

This brought us to a discussion of where we stand on the transition from 
ROOT 5 to 6. Nathan brought up two issues:

 1. When analyzing trees from the analysis launch, ROOT 5 goes about 50%
    faster.
 2. There is an incompatibility between ROOT 6 and the DevTool technique
    for upgrading the compiler on a "RedHat" 6 machine. See his email
    for details
    <https://mailman.jlab.org/pipermail/halld-offline/2016-August/002446.html>.
    This can be finessed by building ROOT 6 with the deprecated
    ./configure-make method.

We decided to make ROOT 6 the default for builds going forward. ROOT 5 
compatibility will continue to be a goal for the standard build, and 
instances of such builds will be maintained at JLab. Paul mentioned that 
he will try to keep ROOT 5 compatibility for the analysis library[?].


      Analysis Launch on sim1.1

Alex completed this last week, as he had promised. See his announcement 
<https://mailman.jlab.org/pipermail/halld-physics/2016-August/000749.html> 
for the details.


      Next Reconstruction Launch

Paul gave the report:

  * The original deadline for code changes was today. It has been moved
    forward to Friday.
  * Will include Will McGinley's new BCAL non-linearity correction and
    the inclusion of pre-shower energies into REST.
  * Will include Mark Dalton's covariance matrix for BCAL showers if
    ready. They will have to be tested for their effect on kinematic
    fitting.
  * Will try to include tracking changes from Simon. He is looking at
    inefficiencies in certain regions of the tracking space for busy
    events (4πp). He is working on changes to the track finding code.
    The changes have not been pushed to GitHub yet and are still under test.
  * Paul is looking at the cause for asymmetries Reinhard Schumacher has
    reported in Reinhard's anti-proton analysis when he applies a tight
    cut on kinematic fit confidence level. Paul thinks he may have a
    fix, but has not really tried it yet. If this fix is not the root
    cause, we may have to go ahead with the launch anyway.


      Offline EVIO skim improvements

Sean has been working on tools to reduce the size of EVIO skims (raw 
data), and has applied them to the pi0 skims for the FCAL and BCAL. The 
user can specify reconstructed objects of interest, and only the raw 
data needed to perform the reconstruction of those objects will be saved 
in the output events. For these pi0 skims, the fully-reconstructed 
DVertex object is also saved so that the event origin is known without 
having to re-do tracking. This results in a reduction of the skim file 
sizes by a factor of 40 to 50. This work was facilitated by use of the 
new EVIO event parser.

[Added in press, from Sean] Will M. just told me [Sean] that with this 
reduced BCAL EVIO pi0 skim, he is seeing a processing rate of 84.5 kHz 
compared to 5.6 Hz with the old pi0 skim, using 4 threads. So, not 
having to do tracking gives us 4 orders of magnitude improvement.


      Change meeting time to 2:00 pm

We decided to move the time of the meeting from 1:30 pm to 2:00 pm, 
following the example of the Online Meeting.


      Stale Branches

Mark reminded us that we should delete branches from the Git repository 
that we do not need anymore. The person who created the branch 
originally is responsible for deleting it when it has served its 
purpose. See the GitHub site for a list of stale branches 
<https://github.com/JeffersonLab/sim-recon/branches/stale>.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20160901/8a54692d/attachment.html>