Re: [netcdfgroup] Inconsistent arguments to collective routine

• To: "Hernan G. Arango" <arango@xxxxxxxxxxxxxxxxxx>
• Subject: Re: [netcdfgroup] Inconsistent arguments to collective routine
• From: Robert Latham <robl@xxxxxxxxxxx>
• Date: Mon, 26 Jan 2009 17:33:08 -0600

On Mon, Jan 26, 2009 at 05:26:49PM -0500, Hernan G. Arango wrote:
> Hi All,
>
> I continue my testing of NetCDF-4 parallel I/O in ROMS.
> It works fine in several different applications but it is
> failing in one with larger grid size. My interface is generic
> so all the applications are using the same sequence of calls
> but with different grid size.
> Since the error is internal it is difficult to follow in the
> debugger.  I haven't been able to reproduce the error in a
> much simpler program.
... snip
>
> I was wonder if anybody has seen this problem before.  This seems
> to be a HDF5 error, perhaps it needs to be send to that group.

someone posted a similar question as you a few months ago.  i can't
link you to it since there are no archives.  the crux of the issue is
that collective I/O was done from one process, which is an error.

possibility #1: you, as a caller of netcdf4, are doing this collective write
from one process (be sure to involve all processes, even if they have
no i/o do to: just make the count array on those processors 0,0,0...)

possiblity 2: you are involving all processes and netcdf4 is being a
little too helpful and no-opping i/o when the count array is 0,0,0... ?

I don't know the answer.  I'm just a spectator :>

==rob

--
Rob Latham
Mathematics and Computer Science Division    A215 0178 EA2D B059 8CDF
Argonne National Lab, IL USA                 B29D F333 664A 4280 315B

• References:
  • [netcdfgroup] Inconsistent arguments to collective routine
    • From: Hernan G. Arango