does parallel IO work with pnetcdf?

[lmkli]

Hello,

I learned that ROMS do the parallel IO with NETCDF4 and HDF5, but for some reason I don't have them in hand so far.

So I want to know if ROMS can do the parallel IO with pnetcdf? If ROMS can do this, how do I modify makefile and *.mk? And, does it need HDF5?

Any help will be appreciated.

M.

[arango]

No, ROMS does not uses the pnetcdf library. This library was developed by the Argonne Lab and it is not a UNIDATA product. It has different library calls. ROMS uses only the NetCDF library distributed by UNIDATA. There are still issues with the efficiency of this library and in some computers the serial library is more efficient than the parallel library.

[gcreager]

netcdf-4.1.1 has "classic" netcdf calls, and can incorporate pnetcdf for parallel IO. What I think you're saying is, ROMS doesn't have a mechanism for invoking MPI-IO, correct?

[arango]

No, it uses parallels calls from the NetCDF-4 and HDF5 libraries which use MPI2 with the MPI I/O layer. See the following post. I have not examined the NetCDF 4.1.1 library. As far as I know, UNIDATA NetCDF does not support neither distribute the pnetcdf library.

[ce107]

I have not examined the NetCDF 4.1.1 library. As far as I know, UNIDATA NetCDF does not support neither distribute the pnetcdf library.

NetCDF 4.1.1 allows pnetcdf to be used when using the classic v3 mode - the configure script has a --with-pnetcdf= option - UNIDATA does not distribute the pnetcdf library so it needs to have been built separately but it sure does work.

[richard.walsh]

All,

So what is the final word the it does or does not use Argonne's PNETCDF?

If I have a build of PNETCDF, and then I build NETCDF with the --with-pnetcdf
flag and the pointer to my Argonne PNETCDF install am I correct in thinking
that I just have to point the ROMS makefile to this build of Argonne-parallel
build of regular NETCDF? What else do I have to do? I remember something about
making sure that my CPP settings were correct for using the parallel IO routines.
Does I then get to use either the parallel or non-parallel version depending on
some ROMS input file option?

I have succcessfully build the code both on an SGI with the Intel compilers
and on a Cray XE6 with the Cray compilers, but only with the serial NETCDF and
my IO seems a bit slow, although I have not run any large cases yet. It there
a HOW-TO this question?

Thanks much ... I will be happen to document exactly what I learn if I succeed
in getting this to work.

Richard Walsh
Parallel Applications and Systems Manager
CUNY HPC Center