[mpich-discuss] mpich3 3.0.4 and binding on four socket nodes

[Kenneth Raffenetti]

There was significant improvement to the process binding features in the 
MPICH 3.1 release. Can you build that version and see if it fixes the issue?

Ken

On 05/07/2014 02:25 PM, Susan A. Schwarz wrote:
> I am using mpich3 3.0.4 built with Intel 13.0 compilers on Centos 6.5. I
> am trying to use the 'bind-to  core' and 'map-by core:2' options to
> mpiexec. I am able to run this successfully on nodes in my cluster where
> there are 2 sockets. However, when I try to run it on a 4 socket node
> with 12 cores per socket, I get the following errors:
>
> [proxy:0:0 at f02] get_nbobjs_by_type
> (./tools/topo/hwloc/topo_hwloc.c:189): assert (nb % x == 0) failed
> [proxy:0:0 at f02] handle_bitmap_binding
> (./tools/topo/hwloc/topo_hwloc.c:450): unable to get number of objects
> [proxy:0:0 at f02] HYDT_topo_hwloc_init
> (./tools/topo/hwloc/topo_hwloc.c:527): error binding with bind "core"
> and map "core:2"
> [proxy:0:0 at f02] HYDT_topo_init (./tools/topo/topo.c:60): unable to
> initialize hwloc
> [proxy:0:0 at f02] launch_procs (./pm/pmiserv/pmip_cb.c:520): unable to
> initialize process topology
> [proxy:0:0 at f02] HYD_pmcd_pmip_control_cmd_cb
> (./pm/pmiserv/pmip_cb.c:893): launch_procs returned error
> [proxy:0:0 at f02] HYDT_dmxu_poll_wait_for_event
> (./tools/demux/demux_poll.c:77): callback returned error status
> [proxy:0:0 at f02] main (./pm/pmiserv/pmip.c:206): demux engine error
> waiting for event
> [mpiexec at f02] control_cb (./pm/pmiserv/pmiserv_cb.c:202): assert
> (!closed) failed
> [mpiexec at f02] HYDT_dmxu_poll_wait_for_event
> (./tools/demux/demux_poll.c:77): callback returned error status
> [mpiexec at f02] HYD_pmci_wait_for_completion
> (./pm/pmiserv/pmiserv_pmci.c:197): error waiting for event
> [mpiexec at f02] main (./ui/mpich/mpiexec.c:331): process manager error
> waiting for completion
>
> Any idea what could be causing this problem?
>
> Susan Schwarz
> Research Computing
> Dartmouth
>
>
>
> _______________________________________________
> discuss mailing list     discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss