[mvapich-discuss] mvapich with matlab
Devendar Bureddy
bureddy at cse.ohio-state.edu
Thu Jan 31 16:31:51 EST 2013
Hi Brock
What version of mvapich2 you are trying? Did you try with mpirun_rsh
launcher to see if there is any difference in behavior? Can you also
try with run-time parameter MV2_ON_DEMAND_THRESHOLD, setting it equal
to your job size?
-Devendar
On Thu, Jan 31, 2013 at 4:10 PM, Brock Palen <brockp at umich.edu> wrote:
> In the past I have had a lot of luck using mvapich as a dropin mpich replacement with matlab to enable matlab parallel computing toolbox to use our IB fabric.
>
> This is still working, but recently we have been having errors, when trying to start matlab jobs on more than 64 cores.
>
> Does the wireup of mvapich do anything unique above 64 ranks? We are using hydra as our launcher and we don't have this problem, using the same hydra install with mpich-1.5.1 on the same nodes/matlab install.
>
> Also this used to work and am at a loss trying to discover why it has changed.
>
> <snip>
> [proxy:0:0 at nyx5043.engin.umich.edu] HYDT_dmxu_poll_wait_for_event (./tools/demux/demux_poll.c:77): callback returned error status
> [proxy:[proxy:0:2 at nyx5048.engin.umich.edu] HYD_pmcd_pmip_control_cmd_cb (./pm/pmiserv/pmip_cb.c:886): assert (!closed) failed
> </snip>
>
> Brock Palen
> www.umich.edu/~brockp
> CAEN Advanced Computing
> brockp at umich.edu
> (734)936-1985
>
>
>
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
--
Devendar
More information about the mvapich-discuss
mailing list