Difference between revisions of "GlueX Offline Meeting, May 18, 2010"

From GlueXWiki
Jump to: navigation, search
(Minutes)
m (Text replacement - "http://argus.phys.uregina.ca/cgi-bin/private" to "https://halldweb.jlab.org/doc-private")
 
(14 intermediate revisions by the same user not shown)
Line 22: Line 22:
 
===Slides===
 
===Slides===
  
Talks can be deposited in the directory <code>/group/halld/www/halldweb1/html/talks/2010-05</code> on the JLab CUE (you have to be a member of the "halld" Unix group). This directory is accessible from the web at https://halldweb1.jlab.org/talks/2010-05/ .
+
Talks can be deposited in the directory <code>/group/halld/www/halldweb/html/talks/2010-05</code> on the JLab CUE (you have to be a member of the "halld" Unix group). This directory is accessible from the web at https://halldweb.jlab.org/talks/2010-05/ .
  
 
=Agenda=
 
=Agenda=
Line 33: Line 33:
 
# [[GlueX-Collaboration-May-2010#Tuesday_May_11.2C_2010|Review of Collaboration Meeting]]: all
 
# [[GlueX-Collaboration-May-2010#Tuesday_May_11.2C_2010|Review of Collaboration Meeting]]: all
 
# GlueX Software Workshop(?): all
 
# GlueX Software Workshop(?): all
# [http://argus.phys.uregina.ca/cgi-bin/private/DocDB/ShowDocument?docid=1530 JLab Scientific Computing Plan]: all
+
# [https://halldweb.jlab.org/doc-private/DocDB/ShowDocument?docid=1530 JLab Scientific Computing Plan]: all
 
#* [https://mailman.jlab.org/pipermail/halld-offline/2010-May/thread.html email thread]
 
#* [https://mailman.jlab.org/pipermail/halld-offline/2010-May/thread.html email thread]
 
# [[HOWTO set up the GlueX environment]]: Mark
 
# [[HOWTO set up the GlueX environment]]: Mark
Line 44: Line 44:
 
'''JLab''': Craig Bookwalter, Mark Ito (chair), David Lawrence, Sascha Somov, Simon Taylor, Elliott Wolin, Beni Zihlmann
 
'''JLab''': Craig Bookwalter, Mark Ito (chair), David Lawrence, Sascha Somov, Simon Taylor, Elliott Wolin, Beni Zihlmann
  
plugin tutorial from dave for a future meeting
+
==Announcements==
  
very likely, another student from Yelena, rootspy candidate
+
# '''New release''' Mark put together a new release, [https://halldsvn.jlab.org/repos/tags/sim-recon-2010-05-17/RELEASE sim-recon-2010-05-17], announced yesterday.
 +
# '''Contact with Dmitri''' David and Mark have been in email contact with Dmitri Romanov about prospective work on a calibration database.
 +
# '''Writing out tracking results''' David led us through his recently published [[HOWTO save tracking results to an HDDM or EVIO file for later playback]].
 +
#* David agreed to give us a tutorial on the use of plugins in the JANA framework at a future meeting.
 +
# '''CNU Student''' Elliott told us that we will likely get another CNU student to work on software this summer via Yelena Prok. He and David are discussing appropriate projects, but ROOTSpy is a strong candidate.
 +
# '''HDF5''' Elliott informed us that Gagik Gavalian, working in Hall B, is interested in using [http://www.hdfgroup.org/HDF5/ HDF5] as a data format for archiving analysis results (including calibration constants). Craig has had some experience with it, using it for a home-brew ntuple format.
  
hdf5
+
==Review of minutes from the last meeting==
  
mantis on the way
+
We went over the minutes from the May 4th meeting.
  
craig appreciated grad student as well as others
+
* Mark reported that the Offline Computing change request has been entered in the official project software.
 +
* Elliott reported that Mantis on the way; just a few more details to iron out.
  
calorimeter cluster issues: indiana student
+
==Review of the Collaboration Meeting==
  
bcal mcsmear for new student
+
* Craig commented that he really appreciated having the graduate student workshop at the meeting. Things were learned. Others concurred.
 +
* Sasha asked about what the next steps on the calorimeter cluster issues are going to be. David reminded him that a student at IU will be working on it. Matt Shepherd has suggested starting him on the migration of BCAL smearing to mcsmear as an initial project.
 +
* Elliott remarked that [http://openmp.org/wp/ OpenMP], mentioned in Chip Watson's talk, was a simple way to achieve parallelization.
  
openMP: simple parallelization
+
==GlueX Software Workshop==
  
software workshop, paul eugenio, jsa initiatives, june or july proposal due, decide in advance, next offline meeting need to have something ready,
+
David described the idea, proposed by Paul Eugenio, that we try to have a Software Workshop using JSA Initiative funds, along the lines of the the [http://conferences.jlab.org/CLAS12Software/ CLAS12 workshop] scheduled for later this month. The proposal is due rather soon (in a month or two). We appointed a sub-committee to come up with a proposal for the next offline meeting, after the CLAS12 workshop occurs. Members are David, Beni, Simon, and Sascha.
Dave Beni Simon Sascha
+
  
dispute over chips numbers
+
==JLab Scientific Computing Plan==
graham does not mention jana, does mention clara
+
gluex computing plan?
+
requirements, srm
+
monte carlo at outside institutions: sascha
+
2002 computing plan
+
computing model
+
  
cross out talk to graham about evio
+
Graham Heyes's recently posted [https://halldweb.jlab.org/doc-private/DocDB/ShowDocument?docid=1530 plan] for the current fiscal year generated a lot of [https://mailman.jlab.org/pipermail/halld-offline/2010-May/thread.html discussion on the offline email list]. We discussed the plan and the discussion.
 +
* We thought that Chip's characterization of data volume at less than 1% of an LHC experiment was probably off the mark, low.
 +
* Graham did not mention JANA in his report in the section on reconstruction tools. This will be remedied.
 +
* Mark asked whether we should revisit our computing plan for GlueX as described in the [https://halldweb.jlab.org/doc-private/DocDB/ShowDocument?docid=58 Hall D Design Report], reviewing the computing model for the experiment. This was thought to be a good idea. This would go a long way to developing a set of requirements.
 +
* Any requirements should certainly include the SRM or equivalent.
 +
* Since offsite computing resources are significant in our current inventory, and will likely remain so, we should develop an infrastructure-appropriate plan to exploit them. Craig pointed out that even now, with his CLAS 6 GeV analysis, he is using the FSU farm. Blake Leverington has used the UConn OSG site for his GlueX thesis work. Sascha mentioned that although the bulk of the reconstruction for Belle was done at KEK, outside institutions contributed cycles to Monte Carlo generation.
  
Beni demo
+
==Setting up the GlueX software environment==
 +
 
 +
Mark led us through his [[HOWTO set up the GlueX environment]] wiki page. The philosophy is to provide a set of override-able defaults for a minimal set of fundamental variables, and then derive all environment variables needed to build and run our software from this set.
 +
 
 +
David objected to this scheme. No consensus was reached on recommendations to the collaboration.
 +
 
 +
==[[Progress toward constructing our own matrix classes]]==
 +
 
 +
Simon presented [[Progress toward constructing our own matrix classes|recent work]] on a set of custom matrix classes with fixed dimensions appropriate to the tracking problem. Improvements come from eliminating checking of array dimensions, unrolling loops (calculating matrix elements directly), and use if SIMD instructions. He achieves increases of speeds of factors of a few for many of the operations he benchmarked.
 +
 
 +
The next step is incorporate use of these classes in the Kalman filter fitter.
 +
 
 +
==Hardware/Connection tracking database==
 +
 
 +
Beni presented a PHP script he had developed to track components and connections for the FDC. It provides an web-based interface for entering information as components are produced and wired up. He received many comments from the group. Among them, Elliott reminded us that the online group has studied the use of [http://www.aps.anl.gov/epics/irmis/ IRMIS] for a similar function.
 +
 
 +
==Action Item Review==
 +
 
 +
We did not get to this. Mark will send out an email to the offline email list with a list of outstanding items.
 +
 
 +
==New Action Items==
 +
 
 +
# Develop a computing plan for GlueX. -> Mark
 +
# Send out list of outstanding action items. -> Mark

Latest revision as of 17:11, 24 February 2017

GlueX Offline Meeting
Tuesday, May 18, 2010
2:00 pm EDT
JLab: CEBAF Center F326/7

Telephone

To connect by telephone:

  1. dial:
    • 800-377-8846 : US
    • 888-276-7715 : Canada
    • 302-709-8424 : International
  2. enter participant code: 77438230# (remember the "#")

Video Conferencing

Slides

Talks can be deposited in the directory /group/halld/www/halldweb/html/talks/2010-05 on the JLab CUE (you have to be a member of the "halld" Unix group). This directory is accessible from the web at https://halldweb.jlab.org/talks/2010-05/ .

Agenda

  1. Announcements
    1. New release, sim-recon-2010-05-17: Mark
    2. Contact with Dmitri: David, Mark
    3. Writing out tracking results: David
  2. Review of minutes from the last meeting: all
  3. Review of Collaboration Meeting: all
  4. GlueX Software Workshop(?): all
  5. JLab Scientific Computing Plan: all
  6. HOWTO set up the GlueX environment: Mark
    • Can we make binaries-in-path a requirement?
  7. Progress toward constructing our own matrix classes: Simon
  8. Action Item Review: all

Minutes

JLab: Craig Bookwalter, Mark Ito (chair), David Lawrence, Sascha Somov, Simon Taylor, Elliott Wolin, Beni Zihlmann

Announcements

  1. New release Mark put together a new release, sim-recon-2010-05-17, announced yesterday.
  2. Contact with Dmitri David and Mark have been in email contact with Dmitri Romanov about prospective work on a calibration database.
  3. Writing out tracking results David led us through his recently published HOWTO save tracking results to an HDDM or EVIO file for later playback.
    • David agreed to give us a tutorial on the use of plugins in the JANA framework at a future meeting.
  4. CNU Student Elliott told us that we will likely get another CNU student to work on software this summer via Yelena Prok. He and David are discussing appropriate projects, but ROOTSpy is a strong candidate.
  5. HDF5 Elliott informed us that Gagik Gavalian, working in Hall B, is interested in using HDF5 as a data format for archiving analysis results (including calibration constants). Craig has had some experience with it, using it for a home-brew ntuple format.

Review of minutes from the last meeting

We went over the minutes from the May 4th meeting.

  • Mark reported that the Offline Computing change request has been entered in the official project software.
  • Elliott reported that Mantis on the way; just a few more details to iron out.

Review of the Collaboration Meeting

  • Craig commented that he really appreciated having the graduate student workshop at the meeting. Things were learned. Others concurred.
  • Sasha asked about what the next steps on the calorimeter cluster issues are going to be. David reminded him that a student at IU will be working on it. Matt Shepherd has suggested starting him on the migration of BCAL smearing to mcsmear as an initial project.
  • Elliott remarked that OpenMP, mentioned in Chip Watson's talk, was a simple way to achieve parallelization.

GlueX Software Workshop

David described the idea, proposed by Paul Eugenio, that we try to have a Software Workshop using JSA Initiative funds, along the lines of the the CLAS12 workshop scheduled for later this month. The proposal is due rather soon (in a month or two). We appointed a sub-committee to come up with a proposal for the next offline meeting, after the CLAS12 workshop occurs. Members are David, Beni, Simon, and Sascha.

JLab Scientific Computing Plan

Graham Heyes's recently posted plan for the current fiscal year generated a lot of discussion on the offline email list. We discussed the plan and the discussion.

  • We thought that Chip's characterization of data volume at less than 1% of an LHC experiment was probably off the mark, low.
  • Graham did not mention JANA in his report in the section on reconstruction tools. This will be remedied.
  • Mark asked whether we should revisit our computing plan for GlueX as described in the Hall D Design Report, reviewing the computing model for the experiment. This was thought to be a good idea. This would go a long way to developing a set of requirements.
  • Any requirements should certainly include the SRM or equivalent.
  • Since offsite computing resources are significant in our current inventory, and will likely remain so, we should develop an infrastructure-appropriate plan to exploit them. Craig pointed out that even now, with his CLAS 6 GeV analysis, he is using the FSU farm. Blake Leverington has used the UConn OSG site for his GlueX thesis work. Sascha mentioned that although the bulk of the reconstruction for Belle was done at KEK, outside institutions contributed cycles to Monte Carlo generation.

Setting up the GlueX software environment

Mark led us through his HOWTO set up the GlueX environment wiki page. The philosophy is to provide a set of override-able defaults for a minimal set of fundamental variables, and then derive all environment variables needed to build and run our software from this set.

David objected to this scheme. No consensus was reached on recommendations to the collaboration.

Progress toward constructing our own matrix classes

Simon presented recent work on a set of custom matrix classes with fixed dimensions appropriate to the tracking problem. Improvements come from eliminating checking of array dimensions, unrolling loops (calculating matrix elements directly), and use if SIMD instructions. He achieves increases of speeds of factors of a few for many of the operations he benchmarked.

The next step is incorporate use of these classes in the Kalman filter fitter.

Hardware/Connection tracking database

Beni presented a PHP script he had developed to track components and connections for the FDC. It provides an web-based interface for entering information as components are produced and wired up. He received many comments from the group. Among them, Elliott reminded us that the online group has studied the use of IRMIS for a similar function.

Action Item Review

We did not get to this. Mark will send out an email to the offline email list with a list of outstanding items.

New Action Items

  1. Develop a computing plan for GlueX. -> Mark
  2. Send out list of outstanding action items. -> Mark