Bug#895748: Default MPI

[Alastair McKinstry]

> Now that mpich builds on all arches, it could be a good opportunity to 
> review and discuss whether we want to keep openmpi as the default
> default, or whether we want to switch to mpich.
>
> Again historically, openmpi had more bugs than mpich, mainly because it
> was trying to do more things (support more arches, for instance).
>
> We seem to have convergence at this point in time.  OpenMPI has now 
> fixed the old egregious bugs, while mpich now supports many arches.
>
> So you could say it doesn't really matter which one we one we choose,
> except perhaps for particular specific HPC configurations, where one
> implementation may have advantage over the other.  But Debian can't
> optimise for that, unless a survey can get back to us on which one
> better suits our actual HPC clients.
>
> Maybe we could swap the default MPI from release to release, to keep 
> them both lively.
>
> Drew
>
I'd favour switching over to mpich for all archs for this release. This
is because MPICH has the failed image support, OpenMPI hasn't.

Fortran 2018 (currently in gcc7 / gcc 8) provides failed image support
for co-arrays, but this requires that open-coarrays is built with mpich.

-- 
Alastair McKinstry, <alastair at sceal.ie>, <mckinstry at debian.org>, https://diaspora.sceal.ie/u/amckinstry
Commander Vimes didn’t like the phrase “The innocent have nothing to fear,”
 believing the innocent had everything to fear, mostly from the guilty but in the longer term
 even more from those who say things like “The innocent have nothing to fear.”
 - T. Pratchett, Snuff