[O-MPI users] Re: [Beowulf] Alternative to MPI ABI

[Jeff Squyres]

On Mar 26, 2005, at 7:50 PM, Greg Lindahl wrote:

>> I guess I don't understand your reluctance to accept a MorphMPI-like
>> solution:
>
> You have repeated your original MorphMPI attributes. I responded to
> them, and I don't see any sign that you've read my response. This is
> not the way discussions are usually held.

Actually, checking the web archives of this thread 
(http://www.open-mpi.org/community/lists/users/2005/03/0028.php), as 
far as I can tell, I have replied to each of your points.  I re-posted 
the MorphMPI attributes because I hadn't seen a direct response from 
you about them.

However, it is possible that I missed your reply; in the web archives, 
I see one offhand comment from you that a MorphMPI-like approach is 
simply an implementation detail, and you saw no problems with it -- is 
this what you were referring to?

If so, are we therefore in agreement that a MorphMPI-like approach is a 
good first step?

I have a followup question:

Who, exactly, wants an MPI ABI?  I have seen a vocal few voice their 
opinions (both for and against).  But these are not representative of 
the overall HPC community.  Have there been any formal surveys 
conducted?  What methodology was used to gather this data?  The reason 
that I ask is because one of the criteria that any software developer 
uses to determine whether to implement something is user demand.  What 
percentage of users actively want or need an MPI ABI?  Who are they, 
and what are *their* reasons why?  Would their needs be solved by a 
MorphMPI approach?  Why or why not?

This, I think, is a critical missing piece to this discussion.  If 
there is a critical mass of users (and more importantly, funders) who 
want/need an MPI ABI, then the whole effort is something that has a 
much better chance of going forward.

-- 
{+} Jeff Squyres
{+} The Open MPI Project
{+} http://www.open-mpi.org/