[Pkg-openmpi-maintainers] Request to Join Project Debian OpenMPI Team

[Tilman Koschnick]

On Fri, 2007-06-15 at 11:29 -0500, Dirk Eddelbuettel wrote:
> On 15 June 2007 at 18:16, oliva.g at na.icar.cnr.it wrote:
> | I Agree. We need to carefully check if the package can cohoexist with
> | the other MPI libraries. I think we should include among these the new
> 
> Yes indeed. At a minimum all run-times needs to be installable at the same
> time.  As for -dev packages, I haven't thought hard about this, but it would
> be need if one could have more than one too.  Longer term goal, maybe?

The alternative system works very well when there is a one-to-one
mapping between two packages, for example a single binary as master and
the corresponding man page as slave.

The problem with the different MPI implementations is that every package
seems to have another set of files under update-alternatives' control,
and within these sets use another master link. Examples are mpiexec,
mpirun, mpicc, mpi.h etc. Then there are differnt ways to split these
files up - everything under the control of a single master, or compile
time and runtime files separately? 

I had problems in the past with differing numbers of slave links as
well. Say package A has mpirun as master and mpirun.1.gz as slave.
Package B only has mpirun as master and no man page. IIRC, if A is
active first and then replaced by B, mpirun points to B, mpirun.1.gz
still to A. Once both packages are removed, mpirun.1.gz remains behind
as a dangling symlink.

So, the one thing we should definitely discuss with the maintainers of
other MPI implementations is the whole alternatives handling in detail,
and not just the priorities to choose. 

> | mpich2 package wich is currently mainteined here:
> | http://torvalds.cs.mtsu.edu/~zach/debian/1.0.2p1
> | by Zach Lowry <zach at zachlowry.net>.
> | What do you think?
> 
> Yes. I had a brief look at MPICH2 as a coworker though the could use it from
> windows. He gave up on that idea, and I am happy with lam for the status quo
> and openmpi as the future, broadly speaking :), but we should coexist nicely
> with the other MPIs.

Since Zach lost interest quite a while ago, and didn't reply to my last
pings, I continued with his packages for my own inhouse use. I never got
to the stage where I could ask for an upload, though, and have switched
to OpenMPI in the meantime. If anyone is interested in more recent -
albeit a bit quick and dirty - MPICH2 packages, I can make them
available.

I think MPICH2 is the one major implementation that didn't join the
OpenMPI group - is that right? So it might very well stay around,
whereas I think MPICH will be replaced by MPICH2 eventually, and LAM by
OpenMPI.

Cheers, Til
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 307 bytes
Desc: This is a digitally signed message part
Url : http://lists.alioth.debian.org/pipermail/pkg-openmpi-maintainers/attachments/20070616/781b7df8/attachment.pgp