[Beowulf] MS HPC... Oh dear...

Joe Landman landman at scalableinformatics.com
Mon Jun 12 08:56:36 PDT 2006

Ashley Pittman wrote:

> I can almost see your argument here, currently there isn't a MPI ABI and
> you think that on Windows this situation might be different?  The

My current belief is that Microsoft will create/mandate the model by
supplying the MPI.  Some standards are defined after the fact.

> technical problems of making a stable, cross ISV ABI are the same on
> Linux and Windows (possibly even slightly worse under Windows), the only
> difference being that under Linux there is a large amount of inertia in
> keeping the status quo as moving to a consistent model would break at
> least somebodies applications, under a brave new Windows world that

Which is fine (breaking of the application), as the benefits IMO far
outweigh the costs.  Simple CBA:

	some apps will break
	some will need to be recompiled, with some minor source code adjustments

	will work with new hardware drivers compiled as .so/.dll linked into
the base (LAM, Scali, and others demonstrate that you can do this).

> inertia isn't there so the ISV might as well mimic the ABI of somebody
> (in this case Microsoft) from day one.

I am not advocating mimicing the Microsoft ABI.  I am advocating getting
a single MPI ABI per ISA ABI.  The question of course is, which one.

> In practise I don't think this is going to be a step forward for two
> reasons, firstly the devil is in the detail so having two different code
> bases export the same ABI might work at the compile/link level but they

Not advocating the ABIs be the same across windows/linux, so I can't
really respond to this well.  The APIs should be the same, but there are
enough platform differences that an ABI being the same across them
doesn't make sense.

Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452 or +1 866 888 3112
cell : +1 734 612 4615

More information about the Beowulf mailing list