Hi Claudiu,
Can you run the job with full debugging options for MPI-Start:
export I2G_MPI_START_DEBUG=1
export I2G_MPI_START_TRACE=1
export I2G_MPI_START_VERBOSE=1
and send me back the standard output and error files? Without those
its quite hard to figure out what the problem could be.
Regards,
Enol.
On Mon, Nov 22, 2010 at 6:52 PM, Claudiu Demian
<[log in to unmask]> wrote:
> Hi,
>
> I've recently noticed that, at our site, the MPI jobs that we submit do
> not communicate between nodes, although they are spread to multiple nodes.
> For example, I've run a simple "Hello world!" program and the only
> output I get is from the last node that was allocated.
> We have both shared /home directory and ssh passwordless
> authentification between nodes.
> The configuration was done according to
> https://twiki.cern.ch/twiki/bin/view/EGEE/MpiTools (Yaim, PBS, MPI-Start).
> What could the problem be and where should I look for it?
>
> Cheers,
> Claudiu Demian
>
|