Software Coordinating Committee Conference Call
			   I/O Subcomittee
		      April 15, 2003 3:00 PM EDT
			  Recorder: C. DeTar

Present:
  Brower, Edwards, DeTar, Holmgren, Mendes, Osborn, Hetrick, Simone, Watson

Absent:
  Mawhinney, Pochinsky, 
  Richards, Petrov, Jung, Gregory, Neff, Genser
  Pochinsky, 

** Action items

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

Agenda: 

     1.) Chip and Don discuss LQCD runtime environment design issues --- topic taken first so Don
         can participate (Read  http://lqcd.fnal.gov/runtime.env )

     Don: See http://lqcd.fnal.gov/runtime.env for a draft.  Please comment.
     Don: What about automatic file migration from the local disk to user disk or 
         should users always do this via job script?
     DeTar: Migration would be useful if the job failed, so the script 
          wouldn't be able to move the files?
     Don: Any opinions about maximum CPU time?  Right now it is 12
     DeTar: 24 would be nicer.
     Don: We could push it up to 24, but would like to hear from Columbia.
     Don: Is it OK to have stdout and stderr inaccessible from the host 
         while a job is running, except by means of a "tail" or "cat" utility?
     DeTar: Seems fine to me.

     Rich: It would be good to have some users gain experience on both clusters
         to see differences in the environment.

     2.)  QDP IO  Interface and Design ---  
    
          Time to finalize the Design and set deadlines on a first release (in May hopefully).
    
     Edwards: The toy QIO model is on CVS at Jlab.  Also a DIME package is
         there.  It has been put under autoconfig and make.  I've
         integrated it with QDP.  Next step would be to solidify the
         XML interface.  Balint and I have been looking at libXML.  As
         for functionality, we distinguish between metadata and
         physics data.  For metadata we can walk up and down the tree
         and make insertions and one doesn't dump the medata until one
         is done.  For physics data, it is done on a fly.
     Brower: Timetable?
     Edwards: May 1 is good because of the upcoming ILDG virtual conference on May 9.
     
     Plenary call concluded at 4 PM EDT.  Edwards, Simone, DeTar, Osborn
     remained for QIO discussion:

     Edwards: I would propose having a MetaReader and PhysicsReader class
       (same for write).  Push and pop operations would be common to both.
       But Meta.. would have more functionality.  MetaReader reads from a string
       PhysicsReader from a file.  Physics data would be large, e.g. correlator
       data, etc. 
     [After some discussion, the group agreed this seems reasonable]
     Edwards:
       What is the API?  Should we allow insertions into a tree?
     DeTar:
       Seems like it would be useful.
     Simone: In QKQCD XML one can have repetitions of a tag: e.g.
       coupling, coupling, ...  This is interpreted as a list.
     Edwards: So how can this be viewed as a directory?
     Simone: The directory approach doesn't work here.  See my libXML
       example.

     [We will continue this discussion during the Apr 17 reserved time.]

     

----------------------------------------------------------------------

Plenary conference concluded at 4:00 PM EDT.  
Next plenary call Apr 24 at 3 PM EDT.