[Halld-offline] Offline Software Meeting Minutes, April 27, 2016
Mark Ito
marki at jlab.org
Fri Apr 29 17:57:17 EDT 2016
Please find the minutes below and at
https://halldweb.jlab.org/wiki/index.php/GlueX_Offline_Meeting,_April_27,_2016#Minutes
_______________
GlueX Offline Meeting, April 27, 2016 Minutes
There is a recording of this meeting <https://bluejeans.com/s/9xG1/> on
the BlueJeans site.
Present:
* *CMU*: Naomi Jarvis, Curtis Meyer, Mike Staib
* *FSU*: Brad Cannon
* *JLab*: Mark Ito (chair), David Lawrence, Paul Mattione, Dmitry
Romanov, Nathan Sparks, Justin Stevens, Simon Taylor, Beni Zihlmann
* *NU*: Sean Dobbs
Announcements
1. *GCC 4.9.2 on farm*. Mark went over his email
<https://mailman.jlab.org/pipermail/halld-offline/2016-April/002307.html>
describing the use of the |module| command to set-up use of GCC
4.9.2 on the farm nodes. The latest versions of our software have
been built with this compiler and installed at JLab.
2. *Analysis Scripts Moved*. Paul led us through his email
<https://mailman.jlab.org/pipermail/halld-offline/2016-April/002310.html>
describing the move of plug-in template generating scripts for use
with the Analysis Library. He received input from Nathan on the
destination location.
3. *New git repository: hd_utilities*. Mark has created a catch-all
GitHub repository for miscellaneous projects. See his email
<https://mailman.jlab.org/pipermail/halld-offline/2016-April/002313.html>
for details.
4. *Bug fix release: sim-recon-1.11.1*. Mark reminded us of the recent
release
<https://mailman.jlab.org/pipermail/halld-offline/2016-April/002314.html>
that fixes a now famous initialization problem that affects tracking
in multi-threaded jobs.
5. *GlueX Software Help Email List*. Mark reviewed his email
<https://mailman.jlab.org/pipermail/halld-offline/2016-April/002297.html>
announcing the list. Collaborators are encouraged to subscribe.
Review of minutes from April 13
We went over the minutes
<https://halldweb.jlab.org/wiki/index.php/GlueX_Offline_Meeting,_April_13,_2016#Minutes>.
* Dmitry reported that the RCDB C++ API has been built and tested on
the ifarm (CentOS 6/5).
* Paul is re-doing his study of multi-threaded behavior of hd_root
with the fix to the famous initialization problem.
Calibration
Calibration Train Processing
Sean brought us up-to-date. He showed a slide
<https://halldweb.jlab.org/wiki/images/5/5e/Sdobbs_OfflineMtg_27Apr16.pdf>.
* The last launch was done with the famous initialization bug (FIB)
present in the code. Sean checked with the experts and the BCAL and
FCAL calibration did not suffer much from this.
* There were a few small problems with jobs.
* He will start soon with calibration of the data from the last two
weeks of the run. He will get the timing calibrations done first.
* There were a few anomalies that arose around the time of the magnet
ramp to 1345 A. In particular the tagger calibrations shifted by 0.5
to 1.0 ns. Also the microscope shows a change in the difference
between FADC and TDC times where the change is a smooth function of
channel. Changes are roughly 3 to 4 ns. This leads to bumps in the
tagger vs. RF timing distributions[?]. Alex Barnes is working on
this issue.
* We noted that the FIB was introduced on March 3.
* Sean is keeping his jobs at eight threads.
* He did some bench-marking on the code after the JANA locking changes
(those that Paul described at the last meeting). He sees a 10-20%
improvement in speed. He did see instances in some calibration
plug-ins where there was extensive calculation being done while
holding on to a lock.
Time-of-Flight Calibration
Beni has done a TOF calibration using the files from the calibration
train. He sees 100 ps resolution per counter over the entire detector
with some counters a bit better. The calibration is done by timing one
plane against another, so for a given counter in one plane comparison
can be done with multiple counters in the other plane. This allows for
an easy visual consistency check of the quality of the calibration.
Spring 2016 Run, Processing Plans
We had a wide-ranging discussion on plans for a comprehensive
reconstruction pass on the recently acquired data.
* Sean reminded us that the plans were discussed at the last
Calibration Meeting. See item (3) of the minutes of that meeting
<https://halldweb.jlab.org/wiki/index.php/April_20,_2016_Calibration#Minutes>.
There is deadline of May 31 for all calibrations to be complete so
we can start the pass.
* An accurate estimate of the computing requirements will come from
statistics for the next Monitoring Launch. We will not attempt to
adjust account priority on the farm until we have those estimates in
hand. That said, Sean thinks that the processing should take about a
month.
* We discussed tagging schemes and tag names for identifying software
and configuration parameters used in the pass. Sean, Justin, Paul,
and Mark all had ideas. We did not come to conclusions and decided
to continue the discussion offline.
* Curtis wondered if we can get away with a smaller number of files
per run for Monitoring Launches. We are now processing the first 20
files of all "is_production" runs[?].
o Sean thought that we might only need 2 to 5 or so. He reported
that this is enough to see ρs and ωs. These signals depends on
many parts of the detector working together properly.
o We noted that the faster the turn-around for the launch, the
faster problems can be found and calibrations adjusted.
o Paul will abort the current launch that has just started. Sean
will suggest to Paul a good number of files to start up with
again. The current calibrations are good enough to learn
something. We do not have to delay further to finalize
calibrations for the last runs of the run.
* We discussed the relative priority of producing REST files vs.
calibration/monitoring in the run up to May 31. There was a
consensus that we should concentrate on the latter.
GlueX TTree DSelector
Paul led us through his email
<https://mailman.jlab.org/pipermail/halld-offline/2016-April/002311.html> describing
a new ROOT selector, DSelector that specialized and improves the
ROOT-supplied TSelector. It is described in detail on its own wiki page
<https://halldweb.jlab.org/wiki/index.php/DSelector>. It is meant to be
used with ROOT trees produced by the Analysis Library and exploits
knowledge of the reaction used to auto-magically provide
reaction-specific operations, including PROOF support, saving work for
the user. Brad has been using it recently and reports an excellent user
experience. [Reminder: Paul pay Brad his $5]
Viewing DEPICSvalue with hd_dump
David described a new feature of hd_dump
<https://mailman.jlab.org/pipermail/halld-offline/2016-April/002312.html> that
can greatly increase the speed of retrieval of EPICS information from
the raw data when the program is run repeatedly. He relies on the native
Linux file-caching mechanism to facilitate retrieval of data from memory
rather than from disk.
Transition to C++11
For the event-parsing work he is doing, David is incorporating a lot of
C++11 language features. He is doing this on a branch, since these
features are not supported by GCC 4.4. They will be supported when we
move to 4.8 on June 1.
He has noticed warning messages compiling our standard code (not his new
stuff) with the new compiler that are not present with GCC 4.4. He was
concerned that other people may be trying to fix these warnings as well
and that would lead to duplication of effort and merge conflicts. There
was not a lot of discussion, and we interpreted this as a lack of effort
by developers to squelch warnings from compilers-of-the-future. We
concluded that no group action is necessary at this time.
Nathan asked about plans for going to ROOT 6. Mark thought that once we
change to the new compiler, that step would be take-able. David thought
we should wait a bit after June 1 so that we are not dealing with two
issues at the same time. We agreed that that would be wise.
RCDB Status
Dmitry updated on the status of RCDB. A lot a work has gone into it
recently. See his slides
<https://docs.google.com/presentation/d/1_uhBIGhZOFy0v41ihW7RXVvonCy2LjrQ81dOAJ3FcFk/edit?usp=sharing>
for full details.
* The RCDB command line is now available. Simple queries can now be
done without having to write a custom Python script. Dmitry showed
some examples.
* An RCDB website is on the way.
* Simple plotting of values is now available. Examples were shown. PDF
files of the plots can be created. Dmitry has a lot of ideas for
future sub-features.
* Custom definitions of "aliases" (set of logical conditions to be
satisfied) and "views" (lists of variables to output) can now be
defined by the user. Definitions are kept in ~/rcdb_user_cfg. Dmitry
showed some examples. So now you can save your go-to queries without
having to type them in each time.
* He gave a list of items for future development.
* He is developing a facility for parsing the configuration files so
that individual parameters can be extracted and turned into standard
RCDB conditions.
o If you have requests for parameters to extract, contact Dmitry.
* Before he leaves (next week, yikes!) he plans to release version
1.0. He listed the features supported in this release.
Retrieved from
"https://halldweb.jlab.org/wiki/index.php?title=GlueX_Offline_Meeting,_April_27,_2016&oldid=74810"
* This page was last modified on 29 April 2016, at 17:55.
--
marki at jlab.org, (757)269-5295
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20160429/89eeb1e3/attachment.html>
More information about the Halld-offline
mailing list