On Feb 8, 2006, at 12:27 PM, rob fiske wrote:
> ==============================================
> Palladium:~/tests/QM/BH4_N fiske$ mpirun C /usr/local/NWChem/bin/
> nwchem
> tests/QM/BH4_N/test.nw
> ----------------------------------------------------------------------
> -------
> It seems that [at least] one of the processes that was started with
> mpirun chose a different RPI than its peers. For example, at least
> the following two processes mismatched in their RPI selections:
>
> MPI_COMM_WORLD rank 0: tcp (v7.0.0)
> MPI_COMM_WORLD rank 2: usysv (v7.1.0)
>
> All MPI processes must choose the same RPI module and version when
> they start. Check your SSI settings and/or the local environment
> variables on each node.
> ----------------------------------------------------------------------
> -------
> ----------------------------------------------------------------------
> -------
> The selected RPI failed to initialize during MPI_INIT. This is a
> fatal error; I must abort.
>
> This occurred on host Cobalt (n1).
> The PID of failed process was 15412 (MPI_COMM_WORLD rank: 2)
> ==============================================
>
> Both machines have LAM-7.0.6 installed, and both are MAC OSX 10.3.9
> for
> their OS and their CPUs are G4's
>
> Has anyone encountered a problem such as this before (I have tried
> giving
> the -ssi option to mpirun as found on this list)?
The error message really does seem to indicate that at least one
process is using LAM 7.1. Since the error message is on rank 2, that
suggests that it might be on the remote node, so you might be running
into path search issues. One quick way to find out is to run "ssh
<node> laminfo" to make sure you are getting the right one. Since
LAM is generally compiled statically by default, make sure you are
running the same version of your code on both nodes - otherwise, the
component lists could be different based on what is compiled into the
LAM library.
Brian
--
Brian Barrett
LAM/MPI developer and all around nice guy
Have a LAM/MPI day: http://www.lam-mpi.org/
|
