[Halld-offline] Offline Software Meeting Minutes, December 7, 2016

Mark Ito marki at jlab.org
Fri Dec 9 13:12:43 EST 2016


Please find the minutes below and at 

   -- Mark


    GlueX Offline Meeting, December 7, 2016, Minutes


  * *CMU*: Curtis Meyer
  * *Glasgow*: Peter Pauli
  * *JLab*: Alexander Austregesilo, Thomas Britton, Mark Ito (chair),
    Nacer Hamdi, Paul Letta, Paul Mattione, Nathan Sparks, Simon Taylor,
    Beni Zihlmann
  * *NU*: Sean Dobbs

There is a recording of this meeting <https://bluejeans.com/s/1ST@C/> on 
the BlueJeans site.


 1. Mark called our attention to the *three releases of sim-recon* since
    the last meeting a month ago:
      * 2.8.0
      * 2.9.0
      * 2.10.0
 2. Flash 250 Emulation
    Sean reminded us about improved fidelity with the firmware and the
    new reporting of errors from compression.
 3. mc_generic CCDB variation
    Sean pointed the existence of this new variation, intended to help
    users who don't know or care about the details of the calibration
    settings for simulated data, for example, those starting to learn
    the system.
 4. *Sim 1.2 test jobs running*. Mark reported that jobs are running
    under both CentOS65 and CentOS7 for this initial test.

      Split of /group and multi-factor authentication (MFA) issues

Paul Letta of CNI gave a presentation 
describing the changes coming up in the new year. The two major things 
to look for:

  * Split of the group disk directories into sensitive (JSA Data) and
    non-sensitive (Open Science Data) file systems.
  * Requirement of an MFA token to access JSA Data.

One general theme (among others) is that, by and large, Lab Users from 
collaborating institutions will not be affected by this change. Another 
is that those working with internal Lab documents for the bulk of their 
work, for example folks in Procurement, will need the token for most of 
their work.

Please see his slides for the details and exact definitions.

      Review of minutes from the last meeting

We looked over the minutes from November 9 
Mark reminded us that the Software Review 
<https://halldweb.jlab.org/wiki/index.php/Software_Review_4> was held 
just after that meeting. It went very well for us; the only 
recommendation was to explore doing data compression in the crates to 
help with bandwidth.

      SciComp Meeting of November 17

Mark flashed his notes from the meeting 
Main message: the CentOS7 farm nodes are up and running now.


Alex A. brought us up to date on the on-going Analysis Launch.

  * He is using the 40 new CentOS7 nodes, each with 72 threads
    available. However he is only able to run 18 threads though he is
    running alone on the node due to memory usage.
  * Why so much memory? There are 100 reactions being analyzed,
    including 40 from Ryan Mitchell's new, comprehensive,
    final-state-survey plug-in.
  * We are now just shy of the 50% point.
  * Paul is working on memory reduction in the analysis library. The
    major culprit are high-multiplicity final states, which generate
    many kinematic fits, each fit producing a large error matrix that
    needs to be saved. One thing he is trying is to save the matrices as
    floats, not doubles.
  * It was noted that there are just a few plug-ins responsible for high
    memory demand.
  * Mark wondered if there are photon quality cuts that can be applied
    to reduce the number of particle combinations tried. The threshold
    is already at 100 MeV which we would not like to increase, but
    perhaps there are other parameters to play with.
  * Alex estimates that "the first part[?]" will be done in 10 days,
    with the second part running over the holiday shutdown.
  * Alex showed us the Launch webpage
    The jobs are running at a steady rate, maintaining 40 instances
    corresponding the the 40 new nodes.

      Geometry Issues in sim-recon

Sean presented ideas he will flesh-out in an upcoming GlueX Note. He 
addressed the issue of how we handle changes in geometry, either due to 
reconfiguration (both major and minor), or re-assessments of positions 
at a fixed point in time (for example development of better alignment 
procedures). Both of these sources of variations are naturally captured 
in the CCDB. His proposal is to (a) control the version of HDDS XML 
files via the CCDB for configuration changes and (b) also control a set 
of offsets, hopefully small, in the CCDB that capture variation from an 
ideal geometry. He also presented the structure for serving out geometry 
information inside JANA. Please see his slides 
for details.

We had some discussion of how we decide which information goes where: 
does some particular change in an offset represent a configuration 
change and thereby requires a new XML file, or simply a run-dependent 
variation from some nominal setting. We will have to develop a 
philosophy on deciding these questions at some point.

      Event Viewer Plans

Thomas is looking into developing a new event viewer for GlueX. He has 
been reviewing features of hdview2 as well as looking at adopting the 
ROOT-based EVE package for rendering a 3D representation of the 
detector. He has already succeeding in generating pictures with EVE 
using the existing ROOT geometry generated by the HDDS package. Dmitry 
Romanov did some work along these lines in the past. Thomas hopes to 
work with him when his visit starts in February.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20161209/ef4e51b9/attachment.html>

More information about the Halld-offline mailing list