Difference between revisions of "GlueX Offline Meeting, September 30, 2015"

From GlueXWiki
Jump to: navigation, search
(Agenda)
 
(2 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
GlueX Offline Software Meeting<br>
 
GlueX Offline Software Meeting<br>
Wednesday, September 30, 2015<br>
+
Wednesday, October 14, 2015<br>
 
1:30 pm EDT<br>
 
1:30 pm EDT<br>
 
JLab: CEBAF Center F326/327
 
JLab: CEBAF Center F326/327
Line 34: Line 34:
  
 
Talks can be deposited in the directory <code>/group/halld/www/halldweb/html/talks/2015</code> on the JLab CUE. This directory is accessible from the web at https://halldweb.jlab.org/talks/2015/ .
 
Talks can be deposited in the directory <code>/group/halld/www/halldweb/html/talks/2015</code> on the JLab CUE. This directory is accessible from the web at https://halldweb.jlab.org/talks/2015/ .
 +
 +
==Minutes==
 +
 +
Present:
 +
* '''CMU''': Curtis Meyer, Mike Staib
 +
* '''FIU''': Mahmoud Kamel
 +
* '''JLab''': Amber Boehnlein,  Mark Ito (chair), David Lawrence, Paul Mattione, Kei Moriya, Eric Pooser, Nathan Sparks, Simon Taylor, Beni Zihlmann
 +
* '''NU''': Sean Dobbs
 +
* '''UConn''': James McIntyre
 +
 +
There is a [https://bluejeans.com/s/8Df9/ recording of this meeting] on the BlueJeans site.
 +
 +
===Announcements===
 +
 +
# '''Team Maintainers and Admins'''. [[GitHub and GlueX|Team Maintainers]] are members of the [https://github.com/orgs/JeffersonLab/teams/gluex "GlueX Team"] that are able to add new members to the team on GitHub. Admins are members of the [https://github.com/orgs/JeffersonLab/teams/gluex-admin "GlueX Admin"] team and have privilege to create and delete repositories owned by the GlueX Team.
 +
#* Currently GlueX Team members are not allowed to create repositories owned by team. There is a setting to allow that privilege. Mark asked whether we should turn that on. David suggested that we keep things as they are; team members have private accounts and can create and delete repositories at will using those accounts. Mature private repositories can be forked into the GlueX Team when they are ready by one of the admins. We agreed with David; keep things as they are.
 +
# '''[[GlueX-Collaboration-Oct-2015|Collaboration Meeting]] October 8-10, 2015 at Jefferson Lab'''. We reviewed the agenda as posted. Paul may need more time than listed, but we decided not to do micro-adjustments to the agenda.
 +
 +
===Review of minutes from September 16===
 +
 +
We went over [[GlueX Offline Meeting, September 16, 2015#Minutes|the minutes]].
 +
 +
* '''New Work Disk'''. The initial rsync of our work disk to the new Lustre server completed last week. It only took about 4 days. The switch-over will require a down period for the final "top-off" rsync of several hours. We will postpone that until after the collaboration meeting.
 +
* Sean has confirmed that '''mailman.jlab.org is now accessible''' from off-site.
 +
 +
===Offline Monitoring===
 +
 +
Kei gave the report. See [https://halldweb1.jlab.org/wiki/images/e/e5/2015-09-30-offline_monitoring.pdf his slides] for details. Topics covered included:
 +
* Current launch (number of jobs, nature of failures)
 +
* Statistics on the launch (memory use, CPU time vs. wall time, etc.)
 +
* Farm environment (other jobs running in competition with the launch)
 +
* Memory Usage on Farm (number of jobs memory limited, not limited by number of job slots)
 +
** Hall D can be memory efficient due to JANA multi-threading
 +
** Other, non-Hall-D jobs may not be as memory efficient
 +
** We may be tape limited in any case. (Paul)
 +
** If an entire node is reserved for Hall D, then our memory efficiency will translate into greater core usage efficiency. (David)
 +
* Moving Forward.
 +
** Due to new plugins, now require 8GB of RAM to run jobs - some pruning of unnecessary histograms would be welcome
 +
** Request by Mike S. that only certain runs be run with -PBCAL:USE_TDC=1, unfortunately bug in script made all runs processed with this option
 +
*** a negative effect on uncalibrated runs
 +
** Started on BCAL cosmics 2015-06 ver03 on September 29 (Tue)
 +
** Next launch is this week, October 2(Fri): no demand for this
 +
** Any need for running over 2014-10 Data?: no requests received
 +
 +
====SWIF Summary Output====
 +
 +
Kei showed us a [https://halldweb.jlab.org/data_monitoring/launch_analysis/tmp/summary_swif_output_offline_monitoring_RunPeriod2015_03_ver15_hd_rawdata.html new webpage] that can be automatically generated to display statistics about each launch. Some of the plots are among those he has been showing in the launch reports, some are new. There is also a listing of all problem jobs.
 +
 +
===Spring 2015 Commissioning Simulations===
 +
 +
Mark reported that the re-do of the [[Spring 2015 Commissioning Simulations]] with new BCAL code is about 80% done. Output files can be found at /volatile/halld/detcom_02_1.
 +
 +
===2016 Commissioning Simulations===
 +
 +
Sean has started putting together a [[Sim1 Conditions|wiki page]] with information about standard conditions to be simulated next.
 +
* Names for projects: something like sim1, sim2,...
 +
* The data sets will not be necessarily tied to a particular real data run, rather they will be generic and can be used for studies under those generic conditions, for example signal studies for particular reactions or rate studies.
 +
* We may need a larger data set than those we have been generating for commissioning simulations recently, perhaps something on the scale of past data challenges. Right now the only large data set we have is from Data Challenge 2, and that is getting kind of old now.
 +
 +
===Auto-Build on Pull Request===
 +
 +
Sean described the recent effort to set up a build of sim-recon triggered by a pull request on GitHub as we discussed at the last meeting.
 +
* He, Nathan, and Mark met with Marty Wise and he has installed some packages on halldweb to support the system.
 +
* The system uses the "gluex" user on GitHub.
 +
* Mark put together the scripts to do the build. Although initiated on halldweb, the build is executed on the ifarm.
 +
* The CGI script on halldweb not only initiates the build but reports the results back to GitHub as a comment on the originating pull request.
 +
* There seems to be an issue having the entire process succeed from GitHub's point of view, possibly because the build takes a while and the system may be timing out. The build itself seems to succeed.
 +
* The system is not active yet, but will be turned on once this last issue is solved.
 +
 +
===Noise Studies===
 +
 +
Sean showed some initial results from a study comparing raw hit distributions between simulated E&M background data and pair spectrometer triggers. See [https://halldweb.jlab.org/wiki/images/b/b9/Sdobbs_OfflineMtg_20150930.pdf his slides] for all of the plots. There are a number of large discrepancies, but at this stage there were a lot of questions about whether the comparisons were fair. It does appear that the pair spectrometer triggers have a large prompt component. If that conclusion holds up, then those data would not be useful for these comparisons. If a random trigger could be set up, that would solve the problem.
 +
 +
Mark remarked that it would be very nice if we could understand the comparison. In particular, it would give us confidence in extrapolations to higher beam rates using simulated data.
 +
 +
===b1pi results review===
 +
 +
We decided to postpone discussing the [[Automatic Tests of GlueX Software|automatic b1pi test]].
 +
 +
===Review of recent pull requests===
 +
 +
We also deferred this item. The idea was to review as a group the [https://github.com/JeffersonLab/sim-recon/pulls?q=is%3Aopen+is%3Apr list of pull requests] initiated since the last meeting.
 +
 +
====Other Related Git Issues====
 +
 +
Another set of topics that were postponed:
 +
* policy on comments when merging
 +
* alternate workflows for submitting pull requests
 +
* using git rebase, should we?

Latest revision as of 13:37, 13 October 2015

GlueX Offline Software Meeting
Wednesday, October 14, 2015
1:30 pm EDT
JLab: CEBAF Center F326/327

Agenda

  1. Announcements
    1. Team Maintainers and Admins (Mark)
    2. Collaboration Meeting October 8-10, 2015 at Jefferson Lab
  2. Review of minutes from September 16 (all)
  3. Geant4 Update (Richard, David)
  4. Offline Monitoring (Kei) SWIF output
  5. Data Challenge 3 update (Mark)
  6. Spring 2015 Commissioning Simulations
  7. Fall 2015 Commissioning Simulations (all)
  8. Auto-Build on Pull Request (Sean)
  9. Noise Studies (Sean)
  10. b1pi results review
  11. Review of recent pull requests
    • comments on merge
    • alternate workflows for submitting pull requests
    • rebasing?
  12. 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: Curtis Meyer, Mike Staib
  • FIU: Mahmoud Kamel
  • JLab: Amber Boehnlein, Mark Ito (chair), David Lawrence, Paul Mattione, Kei Moriya, Eric Pooser, Nathan Sparks, Simon Taylor, Beni Zihlmann
  • NU: Sean Dobbs
  • UConn: James McIntyre

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

Announcements

  1. Team Maintainers and Admins. Team Maintainers are members of the "GlueX Team" that are able to add new members to the team on GitHub. Admins are members of the "GlueX Admin" team and have privilege to create and delete repositories owned by the GlueX Team.
    • Currently GlueX Team members are not allowed to create repositories owned by team. There is a setting to allow that privilege. Mark asked whether we should turn that on. David suggested that we keep things as they are; team members have private accounts and can create and delete repositories at will using those accounts. Mature private repositories can be forked into the GlueX Team when they are ready by one of the admins. We agreed with David; keep things as they are.
  2. Collaboration Meeting October 8-10, 2015 at Jefferson Lab. We reviewed the agenda as posted. Paul may need more time than listed, but we decided not to do micro-adjustments to the agenda.

Review of minutes from September 16

We went over the minutes.

  • New Work Disk. The initial rsync of our work disk to the new Lustre server completed last week. It only took about 4 days. The switch-over will require a down period for the final "top-off" rsync of several hours. We will postpone that until after the collaboration meeting.
  • Sean has confirmed that mailman.jlab.org is now accessible from off-site.

Offline Monitoring

Kei gave the report. See his slides for details. Topics covered included:

  • Current launch (number of jobs, nature of failures)
  • Statistics on the launch (memory use, CPU time vs. wall time, etc.)
  • Farm environment (other jobs running in competition with the launch)
  • Memory Usage on Farm (number of jobs memory limited, not limited by number of job slots)
    • Hall D can be memory efficient due to JANA multi-threading
    • Other, non-Hall-D jobs may not be as memory efficient
    • We may be tape limited in any case. (Paul)
    • If an entire node is reserved for Hall D, then our memory efficiency will translate into greater core usage efficiency. (David)
  • Moving Forward.
    • Due to new plugins, now require 8GB of RAM to run jobs - some pruning of unnecessary histograms would be welcome
    • Request by Mike S. that only certain runs be run with -PBCAL:USE_TDC=1, unfortunately bug in script made all runs processed with this option
      • a negative effect on uncalibrated runs
    • Started on BCAL cosmics 2015-06 ver03 on September 29 (Tue)
    • Next launch is this week, October 2(Fri): no demand for this
    • Any need for running over 2014-10 Data?: no requests received

SWIF Summary Output

Kei showed us a new webpage that can be automatically generated to display statistics about each launch. Some of the plots are among those he has been showing in the launch reports, some are new. There is also a listing of all problem jobs.

Spring 2015 Commissioning Simulations

Mark reported that the re-do of the Spring 2015 Commissioning Simulations with new BCAL code is about 80% done. Output files can be found at /volatile/halld/detcom_02_1.

2016 Commissioning Simulations

Sean has started putting together a wiki page with information about standard conditions to be simulated next.

  • Names for projects: something like sim1, sim2,...
  • The data sets will not be necessarily tied to a particular real data run, rather they will be generic and can be used for studies under those generic conditions, for example signal studies for particular reactions or rate studies.
  • We may need a larger data set than those we have been generating for commissioning simulations recently, perhaps something on the scale of past data challenges. Right now the only large data set we have is from Data Challenge 2, and that is getting kind of old now.

Auto-Build on Pull Request

Sean described the recent effort to set up a build of sim-recon triggered by a pull request on GitHub as we discussed at the last meeting.

  • He, Nathan, and Mark met with Marty Wise and he has installed some packages on halldweb to support the system.
  • The system uses the "gluex" user on GitHub.
  • Mark put together the scripts to do the build. Although initiated on halldweb, the build is executed on the ifarm.
  • The CGI script on halldweb not only initiates the build but reports the results back to GitHub as a comment on the originating pull request.
  • There seems to be an issue having the entire process succeed from GitHub's point of view, possibly because the build takes a while and the system may be timing out. The build itself seems to succeed.
  • The system is not active yet, but will be turned on once this last issue is solved.

Noise Studies

Sean showed some initial results from a study comparing raw hit distributions between simulated E&M background data and pair spectrometer triggers. See his slides for all of the plots. There are a number of large discrepancies, but at this stage there were a lot of questions about whether the comparisons were fair. It does appear that the pair spectrometer triggers have a large prompt component. If that conclusion holds up, then those data would not be useful for these comparisons. If a random trigger could be set up, that would solve the problem.

Mark remarked that it would be very nice if we could understand the comparison. In particular, it would give us confidence in extrapolations to higher beam rates using simulated data.

b1pi results review

We decided to postpone discussing the automatic b1pi test.

Review of recent pull requests

We also deferred this item. The idea was to review as a group the list of pull requests initiated since the last meeting.

Other Related Git Issues

Another set of topics that were postponed:

  • policy on comments when merging
  • alternate workflows for submitting pull requests
  • using git rebase, should we?