Problems running MPI

Sean Dilda agrajag at scyld.com
Fri Aug 10 07:59:10 PDT 2001


On Mon, 06 Aug 2001, german kogan wrote:

> 
> 
> I installed Scyld on my master node and got one test slave node up. I was
> trying to run a simple MPI program. And when I tried to run it
> using the command /usr/mpi-beowulf/bin/mpirun -np 2 a.out I got the
> following error message "p0_2813: p4_error: net_create_slave: host not
> bproc node: -3 p4_error: latest msg from perror: Success". However, it did
> work when I used 1 process instead of 2. Any ideas of  what might be the
> problem be?

I think I finally figured out what might be causing this.  beompi (the
MPI implementation shipped in 27bz-6 and 27bz-7) liked to always put the
first slave node job on node 1, then keep counting up.  You don't have a
node 1, so it is having a problem sending the job to node 1.  As a
solution, turn all your nodes off, then go into beosetup and insert an
'off' node as node 0 and make your real node be node 1 and see if that
solves the problem.

We made the mpich in 27cz-8 actually be smart about putting a job on the
least busy node, so this problem should be gone in that release.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 232 bytes
Desc: not available
Url : http://www.scyld.com/pipermail/beowulf/attachments/20010810/cf9e25da/attachment.bin


More information about the Beowulf mailing list