Software Coordinating Committee Conference Call
			   I/O Subcomittee
		       May 29, 2003 3:00 PM EDT
			  Recorder: C. DeTar

Present:
  Brower, Holmgren, Osborn, DeTar, Simone, Mawhinney

Absent:
  Hetrick, Richards, Jung, Genser, Petrov,  Mendes, Gregory, Edwards, Watson
  Neff, Pochinsky

** Action items

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

  Edwards
    
    ILDG metadata working group wants our opinion about 
    an international standard.
    
  DeTar

    I'd like to start with a discussion of how we intend to use the
    metadata.  What does UKQCD want with it?

  Edwards

    Something close to reproducibility of the gauge configurations.
    No random number information is encoded, however.  So generation
    from first principles is out.
    Generation at the level of physics may be more achievable.
    So coefficients, normalization conventions, etc. are needed.

  Mawhinney

    And the unit of microcanonical time, which varies by sqrt(2).
    This information could be documented somewhere - e.g. in a paper
    and a pointer to a code.

  DeTar

    I would think the goal should be making it possible for someone to
    extend an ensemble.  But why use XML?  It is not well adapted to
    mathematics and algorithms and the ideas behind them.

  Edwards
    
    Searchability is the second goal.  The Tsukuba group wants to
    choose unique names to describe the action, and avoid the lengthy
    detail.

  Mawhinney

    The number of real configuration sets that will be contributed
    is quite limited, so searchability is not a serious consideration.
    And what if someone wants to search on rho, pi masses or r0 or
    some other global parameter, rather than the details of the terms
    in the action?  That would be very useful, but isn't in the
    current schema.

    Archiving propagators may not be as useful as gauge
    configurations.

  Edwards

    We could have a TeX file.  But that is not searchable.  Simplistic
    approach.  What if we recorded only the name, e.g. "Asqtad" along
    the lines of the Tsukuba proposal and a few relevant parameters,
    say quark masses and u0.  Then do a transformation that expands
    the pointer into a tree.  Every configuration contains a pointer
    that identifies the dataset XML.  This could be minimal.

  DeTar

    But the present Asqtad QCDML example seems overly detailed about
    some items and omits other crucial items.  I don't see why someone
    needs to search on all the coefficients of the action when they
    haven't varied over the past four years.  They can be summarized
    by a single searchable word, such as "Asqtad".

  Edwards

    The NERSC model doesn't have enough information to achieve the goals.

  Mawhinney.

    NERSC has Ensemble_ID, sequence_ID and ensemble_label
    The ensemble label could point to a broad XML, TeX, etc. description
    which fulfills the goals.

  Edwards

    That is what we are describing as the "dataset XML".

  Simone

    We have in mind doing one thing with working files and
    another with archivable files.
    Style sheets can be used to do translations from
    private metadata and ILDG metadata.

  DeTar

    Let's propose that the goal should be to ask contributors to
    provide enough information at their option (XML, published
    reference, reference to code, additional "README's") that someone
    might be able to extend ensemble.  But we don't require
    any one way.

  Simone

    We need to make it more concrete.

  Edwards

    Start from the core (required) pieces in NERSC archive.
    Here is how we rearrange them:

  Committee consensus so far...

      file:
        Hdr_version
        Creation_Date
        Ensemble_ID
        Sequence_number (monotonic)
        Dimension

      record:
        Datatype
        Link_trace  (gauge configs only)
        Plaquette   (gauge configs only)

      BinX:
        Endian    (big)
        Precision (either)
 
      record checksum
        checksum

      dataset:
        Ensemble_Label
        Ensemble_ID
        Action
          Boundary_n
        Archive_date
        References
          (Publication)

  Subcommittee:  Mawhinney, DeTar, Edwards, Simone, Osborn to
    continue discussion Monday, June 2 at 3 PM EDT.  Robert
    Edwards to initiate the calls.

----------------------------------------------------------------------

Plenary conference concluded at 3:30 PM EDT.  
Next plenary call June 12 at 3 PM EDT.