[MITgcm-support] MITgcm now uses NetCDF (was: Re: testreport on G4, mnc)

[Ed Hill]

On Sat, 2004-09-25 at 04:16, Yuan Lian wrote:
> Hi, Ed,
> 
> Just curious, will NetCDF be able to deal with outputs from runs with MPI?
> I haven't tried to use NetCDF yet but I prefer to use it for later data
> processing. Thanks!

Hi Yuan,

Yes, the MNC package (the MITgcm--NetCDF interface) works with MPI and
will soon work with multi-threading.  In our testing/verification runs,
you can see the results from the "myrinet" machines which are all using
MPI and are successfully building and running NetCDF examples with
various (g77, PGI, Intel [1]) compilers:

  http://mitgcm.org/testing.html

The NetCDF library itself is not thread-safe so the MNC package does the
I/O in a "per-tile" fashion similar to MDSIO.  However, one of the
benefits of NetCDF over MDSIO is that the latter will produce a file for
every tile and every time step while the former will still produce a
file for every tile but will concatenate multiple time steps within a
file.  This dramatically reduces the number of output files created.

Ed

[1] Please don't be put off by the "fails" reported for the 
  Intel ifc compiler with MPI on our testing pages.  Those 
  runs are being performed with a very aggressive debugging 
  option ("check-bounds") that often fails perfectly acceptable 
  (standards--compliant) Fortran code.  I have confirmed that, 
  without the aggressive debugging options, the Intel compilers 
  do work nicely with MPI and NetCDF.


-- 
Edward H. Hill III, PhD
office:  MIT Dept. of EAPS;  Rm 54-1424;  77 Massachusetts Ave.
             Cambridge, MA 02139-4307
emails:  eh3 at mit.edu                ed at eh3.com
URLs:    http://web.mit.edu/eh3/    http://eh3.com/
phone:   617-253-0098
fax:     617-253-4464