[MPICH] Forking exisiting mpi processes

[James S Perrin]

Greg Lindahl wrote:

> On Wed, Sep 20, 2006 at 02:55:14PM +0100, James S Perrin wrote:
> 
> 
>>	In our particular case the parent is only used to fork a child 
>>	process on demand, it doesn't need to make any MPI calls itself and so the 
>>forked child process "can" assume it's identity.
> 
> 
> As a piece of practical advice, you'll probably find this works in
> some MPIs, it works in more if the parent doesn't call MPI_INIT(), but
> it won't work with all MPIs no matter what, as some intentionally
> disable fork() or do odd things that break in the presence of
> fork(). The standard is so open that implementers do all kinds of odd
> things. And for all you know, the next release of software for
> whatever hardware you've purchased will break your program.
> 
> -- greg
> 

Indeed our first target will be using Fujitsu MPI and I've yet to test it. SGI 
MPI fails though not an intened target. It looks as though we'll have the write 
an MPI spawn based system.

Regards
James
-- 
-----------------------------------------------------------------------------
James S. Perrin,                   | email: james.perrin at manchester.ac.uk
Manchester Visualization Centre,   | www.sve.man.ac.uk/General/Staff/perrin
Kilburn Building, The University,  | tel: +44 161 275 6945
Manchester, England. M13 9PL.      | fax: +44 161 275 6800/6040
-----------------------------------------------------------------------------
"The test of intellect is the refusal to belabour the obvious" -Alfred Bester
-----------------------------------------------------------------------------