Software Coordinating Committee Conference Call
		       January 22, 2004 3:00 PM EST
			  Recorder: C. DeTar

Present:

   Mendes, Brower, DeTar, Osborn, Neff, Jun, Simone, Holmgren, 
   Edwards, Watson

Absent:

   Mawhinney, Gregory, Pochinski, Petrov


** action items
=====================================================================

 1. QCDML - SciDAC needs to respond

    Simone: [Summarized Maynard's proposal]
    DeTar:  I sent comments earlier this week.
      We need to add (1) a definitive reference for the action
      and (2) a way of identifying classes of actions.
**  Simone: Need comments from the committee ASAP.  I would like
      to give the committee the SciDAC response tomorrow.
**  Brower: Could you please circulate a draft of the reply
      before sending it off?

 2. DIME

    DeTar: What are the middleware requirements?
    Watson: Much prefer that middleware manages only files.
      Cataloging operation has to be aware of the heirarchical
      organization of the metadata.  Archive the dataset 
      metadata as a file, just like the lattice files.
      The lower level operation is just file based.
      We could make a higher level QCD-specific fetcher
      that retrieves all files associated.

    DeTar: Since there will not be an international DIME standard,
      I would suggest modifying the DIME format somewhat,
      adapting it to our needs.  For example, abandoning
      chunking and removing some of the extraneous flags
      in the headers.
    Edwards: What if we read and write tar directly?
    DeTar: How would that work?
    Edwards: Perhaps we could adapt the source code
      as a subroutine.
**  DeTar: We need a subcommittee discussion of these issues.
    DeTar: Can we safely assume that files can exceed 2 GB?
    Watson: We can't at JLab, but could if we push.  What is the
     time frame?
    DeTar: Our largest gauge configuration files are currently
     just under 2 GB.  I could imagine that in one year
     we will be breaking the 2 GB barrier.
    Brower: We should poll the UK and Japan about any 2 GB limits.
**  Watson: I will check with the middleware group.

3. Run Time Environment

   Brower: We should probably put off the discussion for now.
   Brower: Any further decisions about Level 3 interface?
   Jun: It looks like the MILC format will be adopted.
** Brower: Please get together with Eric Gregory about this.
   Jun: For now we are concerned only with Asqtad/staggered
     Level 3.
   Brower: That seems fine.  Eventually we expect domain wall as well.
   
4. QMP extensions

   Edwards: We (and Balint) proposed some extensions.
   Brower: Mawhinney promised implementation of the present QMP
     by the All Hands Meeting.
   Watson: We should start discussing the extensions now and possibly
     ratify them at the AHM.
   Edwards: Need 
     QMP_Abort
     QMP_verbose (debugging).
     QMP_Route (communication with all nodes participating.)    
   Osborn: Is QMP_Route temporary or permanent?
   Watson: A permanent option. With QMP_Route message passing would
     not involve interrupts.  But the code could be implemented
     quickly.
   Edwards: We will certainly have the full implementation
     of the present QMP ready for GigE by the All Hands Meeting.
   Jun: Could QMP_Route calls be used to do multiple transfers?
   Watson: Probably not.  This could produce deadlocks.
**    I will circulate a wish list of changes to QMP.
    
5. Software Committee Meeting

** Brower: Creutz needs names of attendees.

6. Status of optimized GigE communication 

   DeTar: Where does it stand?  Is Pochinski working on this still?
   Brower: Pochinski should have something in one month.  He is
     also working on Level 3 domain wall code.
   Edwards: He is strip mining the 5th dimension to get a better
     mapping on SSE registers.

=====================================================================
Committee conference concluded at 4:30 PM EST.  
Next plenary call January 29 at 3 PM EST.