[mpich-discuss] specifying machinefile for MPI_Comm_spawn

[Rajeev Thakur]

I am not sure if a machinefile info key to MPI_Comm_spawn is currently
supported, but new processes should be spawned on hosts in the order in
the machinefile used to start the job. Is that not what you are seeing?
Are you using MPD or hydra?

Rajeev
 

> -----Original Message-----
> From: mpich-discuss-bounces at mcs.anl.gov 
> [mailto:mpich-discuss-bounces at mcs.anl.gov] On Behalf Of Jitendra Kumar
> Sent: Wednesday, April 07, 2010 11:30 AM
> To: mpich-discuss at mcs.anl.gov
> Subject: [mpich-discuss] specifying machinefile for MPI_Comm_spawn
> 
> Hi,
> I would appreciate your help with a doubt using MPI_Comm_spawn.
> 
>   In order to specify the target hosts for spawning new child 
> processes 
> through MPI_Comm_spawn call, what is correct "key" value that 
> needs to 
> be set in MPI_Info.
> 
> Normally I always specify the hosts to start the MPI processes, by 
> passing the argument '-machinefile'
> mpiexec -machinefile mf -n np ./program
> 
> However, passing machinefile key through MPI_Info in MPI_Comm_spawn 
> seems to have no effect.
> MPI_Info_set(hostinfo, "machinefile", "mf")
> 
> The spawned processes are getting started on arbitrary nodes on the 
> cluster.
> 
> I tried to set the key "host" through MPI_Info_set(hostinfo, "host", 
> "node2") and it works. But this way I can specify only single 
> host, and 
> is a problem if I am starting MPI_Comm_spawn with large number of 
> processors and wants to have control on where they are started.
> 
> I believe I am missing the correct key to set for MPI_Info 
> for passing 
> the machinefile. Would be a great help if you can provide 
> some pointers.
> 
> Thanks,
> Jitu
> 
> -- 
> Jitendra Kumar
> Doctoral Student
> Department of Civil, Construction and Environmental Engineering
> North Carolina State University
> Raleigh, NC 27695
> jkumar at ncsu.edu
> http://www4.ncsu.edu/~jkumar
> 
> _______________________________________________
> mpich-discuss mailing list
> mpich-discuss at mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>