Software Coordinating Committee Conference Call
		       May 15, 2001 3:00 PM EDT
		         Amended June 3, 2001
			  Recorder: R. Brower


Present:

Celso, Holmgren, Mawhinney, DeTar, Watson, Edwards, Brower

** = action required

The meeting consisted of a discussion of committee organization and a quick
review of where we stand based on Email notes of each participant.

1.) Message passing.

    Chip discussed his notes on efficient messaging on Myrinet and MPI support.  
    Bob M reported that Creutz had MILC with MPI calls running on QCDSP.  Proof
    of principle with terrible performance however. CPS is working on porting
    CPS code to other platforms. No double precision at present. UKQCD
    runs CPS on other processors: UltraSparc, Alpha, ...

    ** Andrew Pochinsky share his optimized UltarSprac code with
       BNL/Columbia.

    ** Bob M send committee description of QCDOC's communication primitives.

    ** Carleton send committee the 10-12 MPI communication primitives used in
       MILC communication layer.


2.) Data grid

    Chip discussed future goals on Web base interface to data files.
    Particle Physics Data Grid. Technology web centric. Globus.  Drag and drop
    file and web batch system, etc. One year to develop at demo. Operates on a
    web server X509 security.  Same as credit number. Include NSF site? He is
    going to a meeting on Data Grid issues in California next week. Report on 
    this later.


3.) Data storage.

    Need to retrieving lattices and propagators. Compatible with NERSC?
    Suggestion to have beginning of header in ASCII then switch to binary.

    ** Carleton and Greg Kilcup to extend the NERSC Scheme to propagators.
       (Proposal before end of summer.)  Richard call Greg.


4.) MILC developments.

    New release of MILC code version 6, MPI parallel IO wrapper being added. Will
    look at P4 with SSE instruction. MILC targeting Stagger and Jlab targeting
    Wilson. What is the effect of data layout on performance.  Don is considering
    injecting extra latency into the Myrinet to study impact on MILC.
    Jlab and FNAL should collaborate on high efficiency P4 core routines.
    


5.) Optimization of single processor code.

    Robert reported on the optimization on  Alpha QCD library by Andrew Pochinsky. 
    Andrew is developing a new tool for writing low level (site-oriented)
    routines.  This is a natural evolution from his earlier work on the
    optimization of QCD kernels for the Alpha in order to broaden its
    scope to include the P4 and other processors.

    Robert is reproducing Luscher's numbers (upto 10% difference) for CG on the
    P4 using SSE and SSE2 instructions and clever data layout. Consistent with
    SZIN strategy.


    ** Carleton send information on MILC library of routines --- data layout and 
       calling  syntax
    ** Bob send information on  CPS library of routines --- data layout and 
       calling  syntax


Call concluded at 4:30 when the Audio-conference time expired!