[Halld-offline] Offline Software Meeting Minutes, March 4, 2015

[Mark Ito]

Folks,

Please find the minutes below and at

https://halldweb.jlab.org/wiki/index.php/GlueX_Offline_Meeting,_March_4,_2015#Minutes 
.

   -- Mark
______________________________

GlueX Offline Meeting, March 4, 2015, Minutes

Present:

  * *CMU*: Curtis Meyer
  * *FIU* Mahmoud Kamel
  * *JLab*: Mark Ito (chair), David Lawrence, Paul Mattione, Paul
    Mattione, Kei Moriya, Eric Pooser, Nathan Sparks, Justin Stevens,
    Simon Taylor
  * *MEPhI*: Dmitry Romanov
  * *NU*: Sean Dobbs


Announcements

  * *FADC125 upsampling algorithm implemented in emulation mode*. David
    led us through two recent emails to the group (see below). Now FADC
    time and charge information can be chosen between (a) firmware
    supplied and (b) emulated quantities. There was also a change in how
    pedestal information is reported; an average is now reported.
      o message 1 of 2
        <https://mailman.jlab.org/pipermail/halld-offline/2015-February/001954.html>
      o message 2 of 2
        <https://mailman.jlab.org/pipermail/halld-offline/2015-February/001955.html>
  * *CentOS 6.5 added to b1pi test*.
  * *Code analysis*. David reported that Mike Staib has been using some
    Intel-provided tools to analyze our reconstruction code. Mike found
    a race condition in the CCDB package that has been causing crashes
    when running multi-threaded. This has been reported to Dmitry and he
    is working on a fix. The High-Performance Computing group at JLab
    has a license for this software (although Mike did not access the
    tool using that license). We will ask Mike to document his
    experience with the package so others can try it out.
  * *Why we upgraded*. Mark commented on the schedule for our recent
    upgrade of the web and database servers. The idea was to wait until
    after the collaboration meeting, but also to switch as far in
    advance of the Spring run as possible. That put things at the Monday
    before last.


Review of minutes from February 4

We looked at the minutes 
<https://halldweb.jlab.org/wiki/index.php/GlueX_Offline_Meeting,_February_4,_2015#Minutes>. 

Dmitry has been working on the *Run Control Database (RCDB)*. He is 
importing information from Sean's data monitoring database and 
refreshing information from re-parsed CODA log files. This work is 
on-going. He has also released documentation 
<https://halldweb.jlab.org/wiki/index.php/RCDB_conditions_python> for 
the system.

Offline Monitoring Report

Kei described the most recent launch of the offline monitoring jobs. 
Please see his [Media:2015-03-04-offline monitoring.pdf|slides] for 
details. Some take-aways:

  * difference in CPU time compared to last time: CPU time much higher
    compared to wall time
  * version 10 vs. version 09 shows good correlation of CPU times
  * version 11 has much lower CPU time than version 10 (David thought
    this might be due to improvements in CDC plug-in efficiency).


Commissioning-branch-to-trunk migration

Justin discovered the cause of the failure of the commissioning branch 
to successfully reconstruct simulated data: a default of zero B-field. 
Assigning a run number taken during commissioning in bggen solved the 
problem; the CCDB will then respond with the correct field map.
This led to a discussion of how we should handle the calibration 
constants for simulated data. There are two cases:

 1. simulations intended to mimic conditions of already-taken real data
 2. simulations to explore conditions beyond those already achieved


At present there are two degrees of freedom that we have to play with: 
run number and CCDB variation. Ideas discussed included, but were not 
limited to:

  * negative number numbers (i. e. (-1)*real-run-number)
  * run numbers greater than 10^6 designated as simulation
  * run-period specific reserved run numbers (run ranges designated as
    simulation only, data taking would then avoid these run numbers, run
    keep-out zones in correspondence to run periods)
  * run numbers with year encoded in the higher-order digits


In the end we formed a consensus on the following scheme for the two cases:

 1. "mc" variation of CCDB: run numbers indicate the real-data run
    numbers that are being simulated. This variation already exists.
 2. user-named variations in CCDB: run numbers have a user-defined
    meaning, variation name reflecting speculative conditions being
    explored, e. g., "high-intensity", "upgrade-study-5".


EM background mix-ins

We agreed to emphasize the importance of having a random trigger for 
study and/or inclusion of electromagnetic background.

mcsmear execution time

David did some measurements and found that the BCAL simulation is using 
most of the CPU time in mcsmear. This is due to the detailed simulation 
of hits implemented when studying different segmentation schemes in for 
BCAL read-out. For other studies we should be able to get away with a 
less detailed but less CPU-intensive approach.
Retrieved from 
"https://halldweb.jlab.org/wiki/index.php?title=GlueX_Offline_Meeting,_March_4,_2015&oldid=64000"

  * This page was last modified on 5 March 2015, at 13:36.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20150305/b9721fa2/attachment-0002.html>