You need to tell mpirun that your system doesn't have homogeneous nodes:

   --hetero-nodes        Nodes in cluster may differ in topology, so send
                                      the topology back from each node [Default 
= false]


On Aug 22, 2013, at 2:48 PM, Noah Knowles <nknow...@usgs.gov> wrote:

> Hi, newb here, so sorry if this is a dumb question but I haven't found an 
> answer. I am running OpenMPI 1.7.2 on a small Rocks 6.1, Bladecenter H 
> cluster. I am using the bind-to-socket option on nodes with different numbers 
> of cores per socket. For the sample output below, compute-0-2 has two 6-core 
> sockets and compute-0-3 has two 8-core sockets.
> 
> [1,4]<stderr>:[compute-0-2.local:03268] MCW rank 4 bound to socket 0[core 
> 0[hwt 0]], socket 0[core 1[hwt 0]], socket 0[core 2[hwt 0]], socket 0[core 
> 3[hwt 0]], socket 0[core 4[hwt 0]], socket 0[core 5[hwt 0]]: 
> [B/B/B/B/B/B][./././././.]
> [1,5]<stderr>:[compute-0-2.local:03268] MCW rank 5 bound to socket 1[core 
> 6[hwt 0]], socket 1[core 7[hwt 0]], socket 1[core 8[hwt 0]], socket 1[core 
> 9[hwt 0]], socket 1[core 10[hwt 0]], socket 1[core 11[hwt 0]]: 
> [./././././.][B/B/B/B/B/B]
> [1,6]<stderr>:[compute-0-3.local:03816] MCW rank 6 bound to socket 0[core 
> 0[hwt 0]], socket 0[core 1[hwt 0]], socket 0[core 2[hwt 0]], socket 0[core 
> 3[hwt 0]], socket 0[core 4[hwt 0]], socket 0[core 5[hwt 0]]: 
> [B/B/B/B/B/B/./.][./././././././.]
> [1,7]<stderr>:[compute-0-3.local:03816] MCW rank 7 bound to socket 0[core 
> 6[hwt 0]], socket 0[core 7[hwt 0]], socket 1[core 8[hwt 0]], socket 1[core 
> 9[hwt 0]], socket 1[core 10[hwt 0]], socket 1[core 11[hwt 0]]: 
> [././././././B/B][B/B/B/B/./././.]
> 
> Is this behavior intended? Is there any way to cause bind-to-socket to use 
> all cores on a socket for the 6-core AND the 8-core nodes? Or at least to 
> have that last binding not spread across cores on two sockets?
> I've tried a rankfile too, but had errors-- that should probably be a 
> separate thread though.
> 
> Thanks,
> Noah
> _______________________________________________
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users

Reply via email to