Software Coordinating Committee Conference Call
		     April 5, 2007 1:00 PM EDT
			  Recorder: C. DeTar


Present: 

  DeTar, Levkova, Brower, Clark, Pochinski, Zhang, Gottlieb, Basak, 
  Fowler, Joo, Mawhinney, Jung, Osborn, Watson, Scholz


Absent (?? Sorry if I missed you ??):

  Khoriaty, Renner, Holmgren, Efstathiadis, Edwards, Simone

=====================================================================
** Action items

============= Agenda =================

 0. Software posters for SciDAC07 Boston Meeting
      (Rich)

    Brower: We need new presenters.
      I'd like to have a presentation on thermo including LLNL thermo.
      Also a poster showing the PERI tie-in and/or FNAL.
    Fowler: Ying or I can present. I'm registered for the meeting.
    Brower: I'll call separately to recruit presenters.

1. Begin to finalize USQCD Propagator File Format
     (Carleton, Osborn, Bob M, et al)

   Mawhinney: UKQCD/RBC met on Tuesday, there was some preference for
      the 12 record format.    
   DeTar: The 12-record format is only a proposal at the moment
      so it can be modified.
   Mawhinney: We would like to store sources as well.
      Store two different types of sources:
        Scalar field (same for each source color and spin)
        Dirac field
   DeTar: Perhaps we should then distinguish the two cases in the
      file XML.
   Balint: If source is stochastic - save parameters
   Mawhinney: Our preference is to save the whole source, so our
      reconstruction doesn't make assumptions about code versions.
   Mawhinney: We'd also like to store a source restricted to a hyperplane.
   DeTar: That is not part of QIO at the moment, so would require
      some significant changes.  QDP has a concept of subsets of
      the spacetime dimension, but QIO currently assumes fields
      live on the entire space-time volume.
   DeTar:
      Another strategy is to save the source time slice as "global" data
      which would then be broadcast to all nodes upon reading. This
      seems a bit impractical, since it requires a fairly large
      chunk of memory.
   Mawhinney: The fifth dimension is needed for a check of our 
     solutions.  Also for the physics of axial vector currents.
     We'd like to save the full 5D volume at least temporarily.
     Problem is, we often split the fifth dimension.
   Joo: We don't save the fifth dimension at JLab.
   DeTar: QIO handles an array of QDP types, but it assumes all of the
     data for a given site is passed to it at once.
   Brower: You could use a shift to localize all of the data for
     a site.
   Mawhinney: Then what happens when a node has no data to write?
   DeTar: QIO takes a set of layout functions that tell it which node
     has which site.  To read/write a field all nodes count through
     the entire lattice volume.  A node processes only the sites it
     has.  If a node has no sites, it doesn't do anything.
   Jung: Will the node mapping be correct for the QCDOC
     if only some nodes are I/O nodes?
   DeTar: Not with the current QIO, which asks QMP which nodes are
     I/O nodes.  But in the next revision, the user passes in
     the I/O node information and the flexibility will be there
     to allow only some nodes to create files.
   DeTar: I anticipate we will need more than three file formats.
     Let's focus on the most immediate use case needs.  I will iterate
**   by email to sort out the remaining issues.
   DeTar: As for the JLab format, I am not suggesting that the legacy
     files be revised.  But do we find it worthwhile to introduce the
     modified format I suggested?  I so can we ask our colleagues to
     start using it for future propagator files?
   Joo: I can't answer for Robert.

2. Multithreading

   Joo: Cray won't support this until Autumn.  Then there will
     probably be a compute-node Linux kernel to support it.

Committee conference concluded at 1:50 PM EDT.  
Next call Apr 12 at 1:00 PM EDT

======================================================================