Software Coordinating Committee Conference Call
		     March 8, 2007 1:00 PM EST
			  Recorder: C. DeTar


Present: 
  Brower, DeTar, Levkova, Fowler, Pochinski, Gottlieb, Basak, 
  Joo, Watson, Osborn, Zhang, Jung, Alan Porterfield

Absent (?):

  Mawhinney, Clark, Khoriaty,  Edwards, Efstathiadis, 
  Scholz, Renner, Simone, Holmgren

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

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

0. Multicore and cache management

   Fowler: I have Alan Porterfield from RENCI on the line.

   Porterfield: I'm looking at an AMD quad-core chip code named Barcelona
         with shared on-chip cache 2 MB.

   Watson: We are looking at those chips for an upgrade.

   Porterfield: In the MILC code, field major order improves the
         efficiency of cache management.  We are looking at further
         cache management strategies within the FORALLSITES loops.

   DeTar: These optimizations would be important for QLA.  You should
         get together with James Osborn.  They are also useful for
         our multicore planning.  That would be Edwards and Joo at JLab
         and Holmgren and Simone at FNAL.

   Porterfield:

         With multicore, we need to come up with new strategies.  For
         example we might have one core do communication, one do
         prefetching and the other two do computation.  Two
         cores alone might just saturate the memory bandwidth.

   Gottlieb: If we can get things into the L3 cache and not just L2,
         that would be good.

1. Posting benchmark results

   Zhang: Should our performance results be published, or password
          protected?  Are there privacy or security issues?

   DeTar: I think once we have a chance to check that they were
          done fairly, we wouldn't mind publishing them.

   Joo: Peter was unhappy about the Bagel DWF performance numbers in
        my talk that got posted on the BU workshop web site.  We need
        to be careful about this.

2. MILC/QCDOC Level 3 RHMC 

   Jung:  I am pretty close to having the code ready for MILC
          Level 3 RHMC.  It took longer than I expected.to
          assemble all the single precision routines.

NO CONFERENCE CALL NEXT THURS BECAUSE OF BNL DWF WORKSHOP.
NEXT CALL TUESDAY Mar 20 TO PLAN THE ALL-HANDS SOFTWARE PANEL.

Committee conference concluded at 1:30 PM EST.  
Next call TUESDAY Mar 20 at 1:00 PM EST

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