Software Coordinating Committee Conference Call
		       November 18, 2004 3:00 PM EST
			  Recorder: C. DeTar

Present:

  Brower, Osborn, DeTar, Maresca, Eric Blum, Renner, Holmgren, 
  Ying Zhang, Mawhinney, Petrov, Watson, Jung, Pochinsky

Absent:

  Edwards,  Simone,  Neff 

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

#1. Status of level 3 Domain Wall  Inverters. 
    What are their performance and restrictions?
    Is the interface set to call them from MILC/Chroma/CPS code base?

     Brower:    Bob you have a Shamir DWF inverter?
     Mawhinney: Yes.  Peter is working on a native 5d implementation, but
                I don't know the status
     Brower:    What are the restrictions on the inverters?
                What is Robert Edwards using?
     Mawhinney: He is calling our code.  Code from Boyle to Joo to Edwards.


(This is urgent because we are about to have proposals sent to the program committee and we need to know the options.)


#2. What is the schedule for level 3 Asqtad  --- odd dimensions/single precision

    Brower: What is the status?
    Jung:   Depends on how QMP test goes.
            It is the next item on my list.

#3. SciDAC reply to ILDG standard format discussion!   (*#@&(*@_(I!@)

    DeTar:  Summarized the current status of discussion.
            We have urged ILDG to use LIME packaging.
            They have not reached an opinion, yet.

#3.5 Lattice repository:

    Brower: What about a machine at BNL for archiving lattices?

    Blum: A SUN w/ 500 GB is being set up right now.  Security issues
          are being worked out.  DOE Science Grid access, probably
          in Jan.

#4. Main topic:   Don/Chip/Kostya on User Environment (See Don't note attached)

    Holmgren: See

           http://lqcd.fnal.gov/user_environ.html
      
           So far I have tested this only against the MILC code.  The plan
           is to allow portability through a standard set of environmental
           variables.  I would like to test it against Chroma code.
      
           Notice the "qmprun" for launching jobs.

   Petrov: On the QCDOC we would not specify the number of nodes.
           We could ignore the -np argument.
   Watson: How about checking the -np argument against the partition
           that was allocated?

   Petrov: The user should do it.

   Petrov: Output files don't end up in the starting directory.
           We need a standard unix chdir to force stdout to the
           starting directory.  Host-generated stdout goes as usual,
           but node-generated output goes to different location.
           
   Petrov: Host compilers are different from node compilers.  So there
           should be separately compiled SciDAC libraries for host
           code and node code.
   Holmgren: We could have a setup command.
   Watson:   for each target architecture.
   Holmgern: I'll take a stab at the syntax.
   Osborn:   How do we want to handle versioning?
   Holmgren: Our current setup takes both a version and a target.
   Holmgren: We need feedback from each of the sites.
   Petrov:   We use QOS, which has its own environment.
   Watson:   The idea is to find a common mode of access for all
             sites.
   Holmgren: It would be good to get a Chroma user on FNAL
   Brower:   Proposal to the exec com tomorrow for friendly users:
             QCDOC: MILC Renner, Gottlieb;  Chroma: Osborn
             FNAL:  Chroma Richards,  MILC Gottlieb
             JLab:  MILC DeTar, Gottlieb Chroma: Orginos
             These users would be expected to watch for and track
             bugs.

QCDOC 64-node machine

   DeTar:     When will a 64-node machine be available? 
   Mawhinney: Our 64-node machine was just put on general allocation
              web page.  Kostya is running now and has found a couple
              of problems.  Please help.
   Osborn:    Where is the new QOS?
   Petrov:    Chulwoo, will your QMP support the new QOS?
   Mawhinney: When it is ready, Chulwoo, please tell James and Dru.
   Jung:      I still have the non-nearest-neighbor QMP which should
              work?

QMP Technical Question

   Jung:      Should we allow a node to send to itself?
   Osborn:    No.

QMP testing progress

   Zhang:   We need to increase the number of nodes a test code can
            cover.
   Jung:    Yes.
   Zhang:   My student is testing QMP overhead.  Is it a worry?
   Watson:  No.  The overhead should be trivial.
   Zhang:   Anything else I could do?
   Watson:  We have a variety of versions of QMP.  It would be 
            good to profile QMP over VIA on the GigE machine.
   Osborn:  I thought Celso had it in the code at one time.
   Brower:  We should ask Celso where that stands.
   Zhang:   I will be doing some benchmarking on the X1, following
            an SC 2004 conversation with Ken Roche.  Also Altix.
   Osborn:  I did some as well.  With proper compiler flags
            I got a speedup of a factor of 40 in performance.
   Zhang:   I will speak with Morton about adding profiling.
            (After speaking with Celso.)

=====================================================================
Committee conference concluded at 4:30 PM EST.  
Next plenary call December 2 at 3 PM EST.