I have done some work on this using MPICH which is very similar to OpenMPI
The i think this is working as soon as you spawn a process it creates a
intercommunicator and you can comunicate with the corresponding working
using that intercommunicator. If you want to communicate between two workers
then you will have to merge all the processes which communicate into one
intra communicator and pass that particular intracommunicator during
I hope this helps
Nortel Gov. Sol.
>From: Eric Thibodeau <kyron_at_[hidden]>
>Reply-To: Open MPI Users <users_at_[hidden]>
>Subject: Re: [OMPI users] Dynamic COMM_WORLD
>Date: Fri, 7 Jul 2006 14:49:04 -0400
>Yes, I was interested in the example such as:
>The way I see it, there are litteraly two programs, a Manager program and a
>Worker program which is started by the manager (the example doesn't show if
>multiple workers are started by repeated calls to the spawn command). I
>unsderstand that MPI_COMM_WORLD is somewhat static but I am unclear as to
>to communicate with future-spawned process... Is there an MPI_COMM_UNIVERSE
>that we can drop in place of the MPI_COMM_WORLD? It's also not clear who
>would be part of MPI_COMM_PARENT other than the process that spawned the
>Well, I know I have some reading to perform on the whole dynamic/spawning
>process, my query was really to know if OpenMPI was up and ready in that
>field of MPI2.
>Le vendredi 7 juillet 2006 14:09, Edgar Gabriel a écrit :
> > Eric Thibodeau wrote:
> > > Hello all,
> > >
> > > Before I embark on a train that will run out of tracks, I wanted to
> > > get a WFF concerning the spwaning mechanisme in OpenMPI. The intent
> > > is that I would program a "simple" parallel application that would
> > > demonstrate the ability of recent MPI implementations (OpenMPI) to
> > > dynamically ADD or remove processes from the parallel task pool. Of
> > > course, any documentation concerning these new features would also be
> > > greatly appreciated ;)
> > Open MPI supports spawning new processes as defined in the MPI-2
> > specification, see e.g.
> > However, please note, that this does not modify MPI_COMM_WORLD.
> > MPI_COMM_WORLD is a static communicator which can not be changed, even
> > in Open MPI...
> > Thanks
> > Edgar
> > _______________________________________________
> > users mailing list
> > users_at_[hidden]
> > http://www.open-mpi.org/mailman/listinfo.cgi/users
>Neural Bucket Solutions Inc.
>T. (514) 736-1436
>C. (514) 710-0517
>users mailing list