[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 20010724: MPI and NetCDF

Dear Dr. Stantz,

> To: "'address@hidden'" <address@hidden>
> From: "Stantz, Keith M" <address@hidden>
> Subject: mpi and netCDF
> Organization: UCAR/Unidata
> Keywords: 200107241456.f6OEuZ111199

The above message contained the following:

> I appear to be having problems integrating netCDF calls within
> MPI routines (LAM-MPI).  Briefly, a routine extracts sections
> of data from a netCDF file and passes this information to a
> variety of Linux PCs over a LAN.  Results are gathered, combined,
> and written to a file using netCDF routines.  However, after
> creating and defining the first dimension successfully, the second
> call to nc_def_dim crashes (either a segment error within MPI,
> or a core dump if I try this after executing an MPI_Finalize).
> Are you aware of any protocols that should be followed when
> integrating MPI and netCDF?  Any suggestions or comments
> would be helpful.

With the possible exception of the Cray T3E, the current netCDF library
does not support concurrent operation. In particular, the netCDF library
is not thread-safe. If you want multiple, asynchronous access to a
netCDF file, then you must centralize and synchronize access to the file
using some mechanism external to the netCDF library. Such a mechanism is
non-trivial, system-dependent, and (regretably) beyond our scope.

A search of the the netCDF support-email archive using "MPI"
found several emails.  You might try contacting the original authors.
The support email archive is available off the netCDF home page at

    http://www.unidata.ucar.edu/packages/netcdf/index.html

You might also try posting a message to the netCDF user's mailing-list.

> Thanks,
> Keith Stantz
> 
> Keith M. Stantz, Ph.D
> Radiology Department
> Indiana University
> 541 Clinical Drive, Rm 157
> Indianapolis, IN 46202
> phone: (317) 274-0285
> fax: (317) 274-4074

Regards,
Steve Emmerson   <http://www.unidata.ucar.edu>