[MITgcm-support] MPI, I think

[Chris Hill]

Hi Mike,

  The problem is almost certainly that you aren't using MITgcm code.

  The MPI error is a very generic one that happens when one of the 
processes participating in an MPI program goes wrong. For exampl if one 
of the processes dies due to a missing input file, file permissions, 
disk space, floating point problem etc... you would get to see this 
error. Also if things don't start up right then you would get this 
error, for example if not all the the processes get started in the 
beginning of a run you would get this error.

Chris
Mike Spall wrote:
> Hi All,
> 
> I have a question not related to MITgcm, but I have not been
> able to make any progress for a long time now and I thought
> some of you may have some ideas.  Hopefully this is not an
> inappropriate use of this forum.  I am trying to run another
> coupled ocean/atmosphere model (I know, blasphemy!)
> on a linux cluster (Microway, Opteron 248s) with mpi and
> portland group fortran.  The model compiles but I get
> the following run-time error:
> 
> p0_19392:  p4_error: net_recv read:  probable EOF on socket: 1
> rm_l_2_19415: (0.250000) net_send: could not write to fd=6, errno = 9
> rm_l_2_19415:  p4_error: net_send write: -1
> 
> I suspect it has to do with communication between the
> ocean and atmosphere models (which is done using the MCT
> software developed at Argonne Natl Lab).  Any ideas would
> be greatly appreciated.  I did not build the model, but it
> has been run successfully on another linux cluster with pgi
> at Oregon State, so I suspect it has something to do with my
> cluster and, perhaps, my mpi, but this is really beyond what
> I know how to deal with.
> 
> Any suggestions would be greatly appreciated,
> Mike
>