Re: MPICH2 packaging

[Pavan Balaji <balaji@mcs.anl.gov>]

> | > Thank you for the inquiry.  I've somewhat left MPICH for now (focusing
> | > on OpenMPI, which I don't maintain but use), and assigned its
> | > maintenance to the Debian Scientific Computing team.  But I think there
> | > are others very interested in MPICH2, and am copying the debian-science
> | > list to gauge interest.
> | 
> | (Adding Camm Maguire, the LAM maintainer as Cc)
>
> That address has been out of commission for a while; Camm used to work there
> but AFAIK no longer does.  I don't have the replacement address handy though.

Got a bounced email from Camm's address. Cc'ing the newer one.

> | This raises an interesting question: if we package mpich2, couldn't we
> | drop mpich(1) and LAM from Debian? Are there cases where it's more
> | interesting to use mpich v1 or LAM than mpich2 or OpenMPI?
>
> As a former Open MPI co-maintainer:  yes, LAM is to be deprecated one day as
> Open MPI is actively developed whereas LAM is dead. On the other hand, Open
> MPI is available on only a subset of architectures. It's tricky.
>
> That said, getting good MPICH2 in would be super too!

MPICH2 has all features that MPICH-1 has (and more), except for 
heterogeneity support (using it across different architectures 
simultaneously). Though not many (any?) people use that feature, if it's 
not too much work it is still useful to have it in there.

Another important issue is that I noticed several packages that are 
integrated with MPICH-1 in the Debian packages list (scalapack-mpich, 
etc.). So, they might break if we drop out MPICH-1, though it should be 
fairly trivial to get them to use MPICH-2.

In any case, I think the easiest thing to do will be to add in MPICH-2, 
encourage everyone to migrate from MPICH-1 to MPICH-2, and eventually 
drop out MPICH-1 after a few years.

 -- Pavan

--
Pavan Balaji
http://www.mcs.anl.gov/~balaji