openmpi loses output across multiple nodes

Axel akoh... at gmail.com
Wed Jul 23 01:29:05 CEST 2008



On Jul 22, 5:41 pm, Eric Shamay <eric.... at gmail.com> wrote:
> Axel,
>
> Which program (mpi or cp2k) are you asking about verbosity? On both I

cp2k.

> leave the defaults set and both behave as expected when running on a
> single node. I tried the extra '-v' switch, and also using the '-i
> <input file> -o <output>'  also without any luck.

check out the section on output generation in:

http://groups.google.com/group/cp2k/web/cp2k-user-self--support-part-2

if that still does not produce some output for a properly
running job (check out how much of the 100% are actually
cpu time and how much are system or wait), you may have
a miscompiled binary or BLACS.

cheers,
   axel.
> On Jul 22, 5:06 pm, Axel <akoh... at gmail.com> wrote:
>
> > On Jul 22, 4:46 pm, Eric Shamay <eric.... at gmail.com> wrote:
>
> > > Hello all,
>
> > > I've succeeded in compiling the parallel cp2k, and have run the
> > > benchmarks included (as well as my own simulations) to satisfy myself
> > > that indeed all is as it seems. Everything works fine when running on
> > > a single node.
>
> > > When running on 2 processors on a single node I issue:
>
> > > mpirun -np 2 cp2k.popt <input file>
>
> > > all is well. Output looks fine and the cpus are both running at 100%.
>
> > > When trying to run the program on 2 nodes (identical hardware and
> > > software setup) I run into a problem. I have a host file for openmpi
> > > that lists the processors, 2 cpus per node, and the networking seems
> > > to be fine. I issue the command:
>
> > > mpirun -np 4 -machinefile ~/hostfile cp2k.popt <input file>
>
> > > and the 4 processes seem to fire up just fine: 2 processes per node
> > > are running and all 4 processes use 100% of each cpu. However, I don't
> > > have any output from cp2k. It seems as though the head node that runs
> > > the mpirun command just gets all the processors running but can no
> > > longer speak to stdout or do any type of I/O.
>
> > eric,
>
> > how verbose is your output? it may be switched to buffered mode
> > (4k) for performance reasons. try the -i and -o flags for input
> > and output.
>
> > cheers,
> >    axel.
>
> > > I'm not sure if this is a common problem but I'm at a loss as to
> > > searches in the man pages, documentation, and on the forum and on
> > > google, etc. Does anyone know what might be happening?
>
> > > Thanks,
> > > ~Eric


More information about the CP2K-user mailing list