GlueX Offline Meeting, February 1, 2017
GlueX Offline Software Meeting
Wednesday, February 1, 2017
2:00 pm EST
JLab: CEBAF Center F326/327
- 1 Agenda
- 2 Communication Information
- 3 Minutes
- Review of minutes from the last meeting (all)
- A wrapper for signal MC generation (Thomas)
- Collaboration Meeting Agenda
- HDGeant3/4 Update (Richard)
- Launches (Alex A.)
- Calibrations Updates (Sean)
- Sim 1.2 (Sean, Mark)
- Review of recent pull requests (all)
- Review of recent discussion on the Gluex Software Help List.
- Action Item Review
- The BlueJeans meeting number is 968 592 007 .
- Join the Meeting via BlueJeans
Talks can be deposited in the directory
/group/halld/www/halldweb/html/talks/2017 on the JLab CUE. This directory is accessible from the web at https://halldweb.jlab.org/talks/2017/ .
- CMU: Naomi Jarvis, Will McGinley
- FIU: Mahmoud Kamel
- JLab: Thomas Britton, Mark Ito (chair), Peter Pauli, Nathan Sparks, Simon Taylor, Beni Zihlmann
- MIT: Maria Patsyuk
- Regina: Tegan Beattie
- UConn: Richard Jones
- W&M: Justin Stevens
There is a recording of this meeting on the BlueJeans site.
- New releases of hdds and sim-recon: hdds-3.9, sim-recon-2.11.0. The release came out on January 24. See the link for the announcement.
- New SciComp Web Portal. The new site is in beta release now. SciComp is soliciting feedback on the page. There is a "Suggestion" button at the top of the page for your convenience in doing so.
Review of minutes from the last meeting
We went over the minutes from the meeting on January 18.
- Backups of the RCDB database in SQLite. Mark reminded us that we cannot use an SQLite database if it is sitting on a Lustre file server. It must be copied to a non-Lustre disk first. Unfortunately, the backup location, /cache/halld/home/gluex/rcdb_sqlite/, on Lustre so that it will get written to tape automatically. (The directory is part of the write-through cache.)
- Lustre Recovery is complete. They were able to restore 4 of the 5 OSTs that had been corrupted. Files that remain that cannot be accessed will not be coming back.
A wrapper for signal MC generation
Thomas has written a wrapper for generation of signal Monte Carlo data. The script does event generation (genr8), detector simulation (hdgeant), and signal smearing (mcsmear) with the input of a minimal set of parameters. Several people have private approaches to this problem, but this is an attempt to consolidate approaches. Other generators will be added in the future.
For now the script can be found in the sim-recon git repository, in src/programs/Utilities/MCwrapper. We discussed whether we should change it to the hd_utilities repository; the matter will be addressed offline.
Collaboration Meeting Agenda
We looked at the current state of the offline section of the collaboration meeting agenda:
Saturday February 18, 2017
- 9:00 Session IIb (100) --- Analysis/Offline/Calibration - (Organizer: Sean Dobbs) - Chair: -
- 9:00 (25) --- Offline Software Status -- Mark Ito
- 9:20 (25) --- Calibration Status -- Sean Dobbs
- 9:40 (20) --- Offline Data Processing --- Alexander Austregesilo
- 10:00 (20) --- HDGeant4 Update --- Richard Jones
- 10:20 (20) ---
If you have suggestions, contact Mark.
Richard sent out an email yesterday summarizing progress on HDGeant4. He notes in closing that "HDGeant4 is now mature enough to pass into production."
At the meeting he told us that development is done; all essential features have been implemented.
He showed details of the comparisons he has used to both expose issues with HDGeant (HDG3) and drive development of HDGeant4 (HDG4) that rely on the MCTrajectories that give verbose, step-by-step information about particle propagation. He turned off production of secondary particles to facilitate comparison.
The build of HDG4 has fixes to the vanilla Geant4 distribution that fixes bugs. He urged us to take advantage of those fixes (as happens naturally in the build).
He compared the total track length of particles produced in three particles: photons, neutrons, and others. The numbers are comparable in the HDG3/HDG4 comparison, but there is about twice as much total track length in HDG4, due mainly to neutrons being propagated down to lower energy.
The total number of particles was shown in three categories: photons, electrons, and others. There are only about 30% more particles in HDG4, but the proportion of photons increases from 25% for HDG3 to 50% for HDG4 due to lower cut-offs in the photon simulation. Conversely, electrons go from 50% to 20%.
After a lot of work studying where time was being spent in HDG4, Richard has improved performance to the point where HDG4 is now a factor of 4 slower, per event per thread, than HDG3. Very roughly speaking, a factor of 2 comes from lower energy thresholds for tracking and the other factor of 2 is just code speed, i. e., the object-oriented C++ code takes twice as long to do the same thing as FORTRAN.
We could increase the throughput by raising thresholds, but at most that would give is a factor of two, with possible reduction in the fidelity of the simulation. He recommends that we proceed with the standard thresholds, at least to start, given the uncertainty in costs vs. benefits from ignoring physics at low energies.
Richard is interested in receiving problem reports as issues on GitHub. He explained his taxonomy of such reports, listed in order of Richard-will-look-at-it priority.
- Bugs. Things that are obvious problems, e. g. seg faults, things that halt progress, things that are non-physical.
- Artifacts. Observation of behavior that "does not make sense". These could be bugs, but they also might be features.
- Anomalies. Reports of behavior of HDG4 that is not exactly the same as with HDG3.
Richard requires problem reports that include a recipe for reproducing the problem. He has provided a tool, hddm-root, that will make spectra from the raw HDDM output of HDG4 (and HDG3). Spectra produced in this way get around the problem of having identical collections of versions of the software stack. In addition, discussion on the gluex-software email list can be used to refine improve the utility of the report.
The DIRC simulation will include optical photons. Work here is not done but is not the top priority at this time.
For details, see the recording. Richard's presentation starts at the 36:00 mark.
Paul have the report. Since the last meeting Alex Austregesilo announced completion of Monitoring Launch, 2016-10, ver03 and Analysis Launch, 2016-02, ver05. The analysis launch took a long time due to a small number of events in multi-photon channels that took forever to analyze. In these channels the cut on the number of neutral particles had to be tightened from 20 to 14.
Beni has noticed that for the Fall 2016 data, more events are needed to get good timing calibrations for the TOF. We discussed several possible reasons for this but did not come to a conclusion about the cause.