Re: [myrinet] Re: RE: Myrinet Socket Migration?

On Mon, Aug 20, 2001 at 02:18:51PM -0500, Jared Hodge wrote:

> I guess my question wasn't as clear as I thought it was.  What I meant
> by socket migration is not going to a different platform, but stopping a
> running job and starting it again on a different machine.  In other
> words, checkpointing.  Does anyone have any experience with this?

Sure, Condor has a working system that does that for serial, PVM, and
MPI jobs using TCP/IP. You'd have to modify the mpich/gm driver layer
a bit to work with it, though: You need to cleanly shut down gm before
the checkpoint, and restart it at the restart.

g