Re: [Myricom help #25645] MPD with myrinet

Keith,

On Wed, 2004-05-12 at 10:33, Keith Poirier wrote:
> hello,
> 
> Not sure if you guys should be answering this or not, but I thought I
> would give it a shot...
> 
> I have Rev D myrinet cards and I am trying to run some mpi jobs using
> MPD. My setup is:
> 
> Opteron hardware running RH 3.0 2.4.21-9.0.3.ELsmp kernel.
> gm-2.1.1_Linux
> mpich-1.2.5..12
> 
> I have compiled mpich and I can run a simple cpi job using mpirun.ch_gm
> as well as standard mpirun, but when I try using mpirun.mpd I get:
> 
> /opt/mpi/bin/mpirun.mpd -y -np 2 /opt/mpi/examples/cpi
> [0] Error: unable to decode data from 0 !
> [1] Error: unable to decode data from 0 !
> 
> mpd seems to be setup properly - according to mpdtrace
> 
> [root@crs3 root]# /opt/mpi/bin/mpdtrace
> mpdtrace: crs3_33526:  lhs=crs0_47932  rhs=crs2_33333  rhs2=crs0_47932
> gen=1
> mpdtrace: crs0_47932:  lhs=crs2_33333  rhs=crs3_33526  rhs2=crs2_33333
> gen=1
> mpdtrace: crs2_33333:  lhs=crs3_33526  rhs=crs0_47932  rhs2=crs3_33526
> gen=1
> 
> I did compile mpich with the --enable=root as well. Any pointers would
> be appreciated.

I forwarded your question yesterday to one of our software developers
and I'm awaiting his reply.

Yes, we have several customers that use MPD with MPICH-GM.  

Please send us your machinefile you're using to spawn the MPI processes.

What is the "-y" run-time option that you're passing to mpirun.mpd?

The error message that you are seeing is coming from
mpid/ch_gm/gmpi_conf.c (lines 282-289), and I have never seen a customer
encounter this error message before.  I have asked the developer for the
circumstances under which you may see this error.

Susan
-- 

------------------------------------------
Susan Blackford
Member of Technical Staff
Myricom Inc.
------------------------------------------