Software Coordinating Committee Conference Call
		      April 21, 2005 1:00 PM EST
			  Recorder: C. DeTar

Present:

   Brower, DeTar, Maresca, Eric Blum, Simone, Holmgren, Watson,
   Renner, Osborn, Efstathiadis, Zhang, Edwards, Mawhinney, Jung,
   Petrov

Absent:

   
   Pochinsky, 
    

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

Status of US QCDOC Hardware

   Mawhinney: 10 out of 12 racks are populated and being tested by
              physics code.  Chulwoo 2, Kosta 1, Dru 1, Mawhinney 4.
              We are doing some speed stressing with higher frequency
              (440 MHz) clocks. It allows us to weed out bad nodes
              faster.  We run independent processor (loopback) tests
              first and then test with message passing.  We are moving
              the culled nodes out of the production machine boards to
              test machine boards.  They will be run with a slower
              clock.

   Renner:    I started running Chulwoo's script on the rack last night.

QMP Stress Testing

   Edwards:   Chulwoo, in your version of QMP, do you use the
              passthrough feature in global sums?  UKQCD finds
              QMP hangs in 1/10^4 operations.
              I am concerned that there is no QMP stress testing,
              since your Level 3 Asqtad uses SCU instead of QMP for
              its global sums.

   Edwards:   Peter was finding that the payload hardware checksums on
              the SCUs were being reported incorrectly.  So he has to
              run user-level checksums.  Are you checking the hardware
              checksums inside QMP?

   Jung:      Not yet.  I was waiting for Peter's changes.

   Brower:    What code can we use for testing QMP this way?
   DeTar:     We can run the MILC Asqtad inverter, which uses QMP.
   Brower:    We also need to get the Chroma code using.
   Edwards:   We need to make this an integral part of the testing.  It
              is better to have the testing experts develop specialized
              code.
   Brower:    Has Ying run the QMP test suite on the QCDOC?
   Jung:      I don't know if Morton finished the test.
** Brower:    I will check with Ying.

QIO/LIME changes

   DeTar:     I have nearly completed a revision of the file conversion
              code and had to make some minor changes to the API.

Status of MILC Asqtad HMC Production code on 1K QCDOC Rack.

   Renner:    I am still tracking the memory problem.  We still don't
              understand where all the memory is going.

   Edwards:   Peter Boyle has a new version of QOS that boosts memory
              from 32 to 49 MB.  This was a fairly simple change.
              Improvements beyond that will require considerable effort.

   DeTar:     When is the SciDAC software going to be installed in
              locations that can be referred to with the environment
              variables specified in the SciDAC run time environment?

   Mawhinney: You should tell Stratos what variables need to be
              defined and instructions to build the SciDAC package
              and he will do it.
** DeTar:     OK

   Edwards:   Chulwoo, do you have the bin script with the MPI version
              of QMP?
   Jung:      Not yet.
              	     

Status of Chroma DW HMC Production code  on 1K QCDOC Rack.

    Osborn:  I had problems getting the 64-node machine booted.

Benchmarking
 
   Zhang:    Dan Reed wants a student to do some comparisons of
             MILC code preformance with and without SciDAC software.
   Watson:   Note that QMP is just a thin layer on top of MPI.
             So QMP vs MPI is not very meaningful.
   DeTar:    If you want to compare QDP with nonQDP MILC, I can
             help with that.

Software Meeting:

   Brower:   We will meet Thursday and part of Friday after the AHM at
             JLab.
   

=====================================================================
Committee conference concluded at 1:45 PM EST.  
Next plenary call April 28 at 1:00 PM EST