Newsgroups: comp.parallel.mpi
From: Nick Nevin <nevin@osc.edu>
Subject: Re: MPIL_Signal
Organization: Ohio Supercomputer Center
Date: 22 Jul 1996 14:02:04 -0400
Message-ID: <vn2afwsdv4z.fsf@alex.osc.edu>


    > Franz-Josef Markus <markus@sylt.iti.mu-luebeck.de> writes:

    > I suggest that rank 0 belongs to the father.
    > Try 
    >   MPIL_Signal(childcomm, 1, LAM_SIGUDIE);
    > to kill the child process.

You are correct in suggesting that rank 0 belongs to the father. The
father is rank 0 in the local group of childcomm. The child however is
also of rank 0 but in the remote group of childcomm.

We omitted to specify accurately the behaviour of MPIL_Signal when the
communicator argument is an intercommunicator. To remedy this it has
been decided that the process of the specified rank in the *remote*
group will be signalled.

This means that processes in the local group of an intercommunicator
must be signalled using a different communicator. MPI_COMM_WORLD should
usually suffice.

I have just released patch 15 to LAM 6.0 on the LAM mailing list
lam-chops. The patch implements the new behaviour.

To subscribe to the mailing list, send a message to
majordomo@tbag.osc.edu with a blank subject and the following line in
the body of the message: 

subscribe lam-chops
                    

The full set of patches to LAM 6.0 can also be obtained from
ftp://ftp.osc.edu/pub/lam/lam60-patch.tar

---nick.

-=-
Nick Nevin				nevin@osc.edu
Ohio Supercomputer Center		http://www.osc.edu/lam.html

