Re: dynamic counter-proposal

William C. Saphir (wcs@nas.nasa.gov)
Fri, 10 May 1996 11:19:04 -0700

Rusty,
Can you provide some details on the reasons you think
the dynamic counter-proposal is so different from the
current one? Sure, it looks different, but the functionality
is almost exactly the same (intentionally), and, despite your
suggestions to the contrary, I think the implementation will be
only slightly different. In a previous note I explained in
detail why the people who generated the proposal (including
several implementors) thought there was very little difference. Yet
you say "major change to MPI_Spawn", "messing up the core functionality",
"essentially starting over". The only argument I have seen
to support this is that the old spawn() becomes spawn+attach.
If this is a serious problem then a very simple convenience
function can fix it. On the implementors side I have yet to
hear any specific argument.

I don't disagree with your historical observations, but I
draw a different conclusion from them. We have spent many
months deciding what functionality we want and understanding
the issues in great detail. The draft of the last meeting
reflects this historical process and is (to some) a confused
jumble of functions, each one targeted towards a different
piece of functionality. It is only now, after having gone
through the whole process that we are able to step back
and make sense of what has been produced. The counter-proposal
does that. It is not a departure, but a reorganization designed
to clarify and unify.

> Those who want to reduce the number of functions should vote out the functions
> that supply capability for spawning independent and non-MPI processes, or vote
> out the ability to spawn multiple executables, or vote out MPI_Spawn as a
> convenience function for the special case of MPI_Spawn_multiple. I'll vote
> for them, but I don't mind if I lose on these issues.

If (and only if) there is strong support for the counter-proposal,
proceeding with votes on the current proposal would be a waste
of time. It is appropriate if the debate is over functionality,
but the counter-proposal provides the same functionality
with a very different set of functions. As someone who has
criticized MPI-1 for its bloat, I want to make sure that
we give ourselves a chance to step back and see the big picture
before locking the historical record into a standard.

Bill