[Halld-offline] Offline Software Meeting Minutes, May 31, 2017

[Mark Ito]

Please find the minutes from the meeting on May 31 below and in their 
usual location 
<https://halldweb.jlab.org/wiki/index.php/GlueX_Offline_Meeting,_May_31,_2017#Minutes>.

_______________________________

Present:

  * *CMU*: Naomi Jarvis, Curtis Meyer
  * *FIU*: Mahmoud Kamel
  * *JLab*: Alex Austregesilo, Amber Boehnlein, Thomas Britton, Brad
    Cannon, Eugene Chudakov, Hovanes Egiyan, Sergey Furletov, Mark Ito
    (chair), David Lawrence, Dmitry Romanov, Simon Taylor, Beni Zihlmann
  * *NU*: Sean Dobbs

There is a recording of this meeting <https://bluejeans.com/s/co6tm/> on 
the BlueJeans site. Use your JLab credential to access it.


      Announcements

 1. AmpTools moved to GitHub
    <https://mailman.jlab.org/pipermail/halld-offline/2017-May/002753.html>
    from SourceForge.
 2. MCwrapper 1.6
    <https://mailman.jlab.org/pipermail/halld-offline/2017-May/002766.html>
    and 1.7
    <https://groups.google.com/forum/#%21topic/gluex-software/3Mmm_vI0VT8>
    Thomas has added PBS and Condor support.
 3. "New" packages in build_scripts: AmpTools, HDGeant4,
    gluex_root_analysis, hd_utilities.
 4. Certificates and GitHub
    <https://mailman.jlab.org/pipermail/halld-offline/2017-May/002768.html>.
    The JLab certificates is no longer needed to communicate with GitHub.
 5. REST production 2016-02 ver04 done
    <https://halldweb.jlab.org/data_monitoring/recon/summary_swif_output_recon_2016-02_ver04_batch01.html>.


      Review of minutes from the last meeting---OSG Progress

We went over the minutes from the meeting on April 19 
<https://halldweb.jlab.org/wiki/index.php/GlueX_Offline_Meeting,_April_19,_2017#Minutes>. 
Since then there has been slow and steady progress on GlueX use of the OSG.

  * Work on the submit host at JLab is progressing. Problems are being
    discovered and solved.
  * OSG personnel are installing software to deploy the CMU cluster as a
    Hosted Compute Element, making it available for our use via grid job
    submissions.
  * A decision on how to deploy the GlueX software stack has been made:
    we will be using Richard Jones' Gridmake system.


      Stale Branch Policy

Mark proposed a policy to reduce the number of stale branches 
<https://github.com/JeffersonLab/sim-recon/branches/stale> in the 
sim-recon repository. The policy is

    Branches that have (a) no commits not present on the master branch
    and (2) have not had any updates in six months will be deleted
    "automatically".

There were no objections to this policy and it was adopted.


      3D Event Display

Dmitry and Thomas have been working on a three-dimensional event display 
for GlueX. The system is based on the EVE package of ROOT and uses the 
JANA framework to retrieve events.

  * Thomas gave a live demonstration (starting at 22:00 in the recording).
  * Dmitry showed slides commenting on the effort
    <https://docs.google.com/presentation/d/1LsaelWTnzkBUzuEs7yHPGq7nzwQq0aVH9VlemOZqQZ8/edit?usp=sharing>
    thus far. Please see his presentation for full details.

Dmitry spent a fair amount of time looking at other software packages 
that address this problem space. Turns out there is not much out there 
that is written to serve as a toolkit, i. e., most event display 
programs are highly detector specific.

So far a lot of the work has focused on solving technical problems just 
to get a working proof-of-principle system. One of the difficulties was 
getting two multi-threaded applications, ROOT and JANA, to work 
together. Not all detectors have been implemented and those that can be 
displayed are treated only at a bare-bones level. One challenge going 
forward will be manpower; there are a lot of detailed decisions that 
need to be made, detector-system-by-detector-system, to get a useful 
product. Hopefully each detector group can contribute, but there is a 
learning curve to be climbed for each contributor. Thomas and Dmitry are 
thinking of ways to reduce the slope.


      Bench-marking on KNL

David presented slides on benchmark tests 
<https://halldweb.jlab.org/wiki/images/d/de/20170531_KNL_benchmark.pdf> 
he performed with our reconstruction on a Knights Landing (KNL) machine. 
Please see his slides for the details. The motivation is to develop 
numbers for proposals to run on new supercomputer, Cori 
<http://www.nersc.gov/users/computational-systems/cori/NERSC%27s>.

Although the comparison is a bit tricky, on a per core basis, on KNL our 
code runs about 20% of the speed we get on a 2016-era Broadwell 
processor. This does not necessarily disqualify us for time on Cori; 
software from other projects are likely similarly challenged.

-- 
Mark Ito, marki at jlab.org, (757)269-5295

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20170602/7c52623d/attachment.html>