GlueX Offline Meeting, November 11, 2015

From GlueXWiki
Jump to: navigation, search

GlueX Offline Software Meeting
Wednesday, November 11, 2015
1:30 pm EST
JLab: CEBAF Center F326/327

Agenda

  1. Announcements
    1. New work disk: /work/halld2
    2. Private Wiki open for business
  2. Review of minutes from October 28 (all)
  3. Offline Monitoring (Kei)
  4. Geant4 Update (Richard, David)
  5. Data Challenge 3 update (Mark)
  6. Future Commissioning Simulations (all)
  7. Binary Distributions of GlueX Software (Nathan)
  8. b1pi results review
  9. Review of recent pull requests
  10. Action Item Review

Communication Information

Remote Connection

Slides

Talks can be deposited in the directory /group/halld/www/halldweb/html/talks/2015 on the JLab CUE. This directory is accessible from the web at https://halldweb.jlab.org/talks/2015/ .

Minutes

Present:

  • CMU: Mike Staib
  • JLab: Mark Ito (chair), David Lawrence, Paul Mattione, Kei Moriya, Dmitry Romanov, Nathan Sparks, Simon Taylor, Beni Zihlmann
  • NU: Sean Dobbs
  • UConn: James McIntyre

There is a recording of this meeting on the BlueJeans site.

Announcements

Offline Monitoring

Kei gave the report. Version 17 was started last Friday, but after it was done he discovered that the monitoring plugins had not been included in the jobs. So on Saturday, Kei launched version 18, correcting the omission. It only took 30 hours to run, but the reasons for better performance are not known at present. From his slides, his concluding bullets were:

  • See much less resource use for both RAM and time, still investigating
  • Studies ongoing of time/resources against #cores/threads shown at collaboration meeting with improvements in tracking
  • No updates from SciComp on hogging all cores on a single node
  • Next launch: November 20 (Fri)

Geant4 Update

David gave us an update. He recently tried to download and build HDGeant4 from the Git repository, but linking failed (problem with threaddb[?]). The Geant4 cmake system adds opacity.

Geant4 has been installed on the Gluon Cluster. The CPP simulation has been built against this version and runs albeit without multi-threading presently.

recon2mc Plugin

David submitted a pull request for a new plugin. From his submission:

recon2mc plugin can be used for writing reconstructed track parameters into an HDDM file that can then be used as input to hdgeant and CPPsim. This is to allow one to easily simulate a sample of tracks with a similar kinematic distribution to real data.

Data Challenge 3

Mark gave an update on Data Challenge 3.

The problem with failed reconstruction of simulated data persisted with a recent version of the software. Sean came through with the correct command line options; problem solved. Even though the old software version might work with the new options, Mark will press on with the more recent code and generate new fake data.

There was a SWIF problem with deleting and re-creating workflows that has stopped progress. Chris Larrieu is working on a fix.

Future Simulations

Sean gave the update.

  • Mark ran a few pilot jobs for Sim1. They look fine. Some of the plugins had hard-wired beam energy cuts not appropriate for 12 GeV running; these will need adjustment.
  • Decided to generate events between 4 and 12 GeV. The lower limit was lowered to account for low energy events leaking up into the data sample.
  • Plan is to generate about 2 millions events to start with, look at those, then generate about one day of day of data, about 1.5 billion events.
  • We will begin concentrating on standard beam photon rates, 1 × 107/s.
  • Richard Jones has a web tools for calculating the correct BGRATE parameter, but the site is down right now.
  • We are thinking of adding some information into the REST file with values that are used for the trigger decision, i. e., energy sums. That would allow studies of trigger thresholds. This will have to be developed.

Calibration Challenge

Sean reported on the plan for a Calibration Challenge. We would start with neutral values for calibration constants (without channel-to-channel variation) and see if we can generate constants in an automated, multi-pass process. Code will have to be modified to fit into this paradigm, e. g., using histograms rather than generating a heterogeneous set of ROOT trees. There is multi-stage feature of SWIF that we may be able to exploit.

Binary Distributions of GlueX Software

Nathan gave a report on a project he has been working on to produce pre-compiled versions of GlueX Software for use on computers that cannot access the public builds at JLab. He is using Docker containers to create binaries for a variety of Linuxes. Three slides from his presentation:

  1. Motivation
    1. Provide precompiled dependencies to build development code against in size amenable to transport
      • Not needed when developing on JLab CUE, since halld group builds are already provided there
      • Useful instead for users working offsite or at home
    2. Provide prebuilt code that is ready to be deployed into production/Farm or utilized offsite for high-level analysis
    3. Facilitate comparing behavior/results of programs across Git commits (keep something like the latest 10 commits)
  2. Installing Binary Distribution
    • hdpm command-line-interface
      1. hdpm fetch-dist
        • install latest available binaries for your platform
      2. hdpm fetch-dist [commit hash]
        • install binaries for particular sim-recon commit
      3. hdpm fetch-dist [url or path]
    • Source the environment setup script
  3. Outlook
    • Package sim-recon for OS X
    • Requires some additional dependencies
    • Other potential Docker applications
    • Deploying sim-recon to cloud for simulation workloads

b1pi results review

We looked through some recent and some past results from the b1pi tests. There was a lot of discussion about the various features of the plots and which changes in the software might have caused shifts in those features. See the recording, starting at the 54:00 mark.

One conclusion of the discussion is that we might want to output results in a more digestible format, and make it easier to compare across different runs of the tests. Using the offline monitoring webpages are an option.

Review of recent pull requests

We reviewed pull requests submitted since the last meeting. See the recording starting at the 1:11:00 mark for the full discussion. We looked at: