parallel netCDF

[John Tannahill]

A while back, I think there was some discussion about the order of
dimensions with regards to C/Fortran.  In my original Fortran test
code, I just mimicked the way things were done in NERSC's csnap.c
code.  That is, in my Fortran code, I just used the same dimensioning
order as the C code.  The Fortran code results then matched the C
code results.

Now I have written two additional Fortran codes.  One that mimics
the way our Fortran modeling code currently does its netCDF I/O,
which uses serial netCDF, the slaves read their portions of the
large arrays, and the slaves communicate their array parts back
to the Master for output, generally in some sort of round-robin
fashion because of memory limitations on the Master.  Anyway, this
test code is up and running.  Note also that this code uses
Fortran ordering for everything and netCDF takes care of
translating to the C ordering that it needs.

The Fortran code that I am currently working on is using your
parallel-netCDF library to replace the serial netCDF.  I do not have
it working yet, but one issue that I have already run into is that
I don't want to artificially "C order" things in this Fortran code,
which I believe I currently will have to do.

Anyway, I wanted to give you a heads up, as I believe this topic
was under discussion a while back.  I do believe that your
parallel-netcdf library will need to handle the conversion to
C ordering itself and not have the application writer have to
deal with it.  I hope this all makes sense, and note that I am
down to my last couple of weeks that I will be able to devote
much time to this work.  I am hoping that a new release will show
up shortly?

I am currently using version 0.8.9 on NERSC's IBM SP (seaborg).

Regards,
John

-- 
============================
John R. Tannahill
Lawrence Livermore Nat. Lab.
P.O. Box 808, M/S L-103
Livermore, CA  94551
925-423-3514
Fax:  925-423-4908
============================