Replacing LAM with OpenMPI in Fedora Core

[Ed Hill]

On Thu, 2006-01-05 at 11:59 -0700, Orion Poplawski wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Ed Hill wrote:
> > 
> > I understand that OpenMPI is supposed to [at least, for some people's
> > perception! ;-)] become the "one true MPI" implementation that eclipses
> > all others due to its very cool new modular design and other features.
> > But, even so, I think it would be a good idea to have and maintain other
> > MPI versions (such as MPICH v1 & v2, LAM, etc.) in Extras so that people
> > have some flexibility.  And to do that, we'll very likely need to setup
> > the multiple MPI packages using alternatives.
> 
> While we are thinking about this, it would be good to be able to support
> multiple versions of a particular MPI compiled with different compilers.
>  Locally I maintain LAM compiled with PGF90 and IFORT and it would be
> nice to be able to have all three installed simultaneously.

Hi Orion,

Yes, we also have groups of machines where we maintain combinations of
compilers, MPI libs, etc.:

  MPI:  LAM, mpich, mpich-vmi, etc.
  Compilers:  GNU (multiple versions), Intel, PGI, etc.

and we do it with an install framework that looks something like:

  /opt/pkg/${PKG_NAME}   and/or   /usr/local/pkg/${PKG_NAME}

and then let users dynamically chose which packages or combinations of
packages to use with the environment modules scripts:

  http://modules.sourceforge.net/

I like the above setup because:

  + its quite flexible and can handle dependencies between 
      the packages pretty gracefully
  + it can be made to work (play nicely!) with the in-Core MPI 
      setup and/or any number of additional MPI setups which 
      might be installed (perhas someday?) through Fedora Extras 
      or local installs
  + its an increasingly popular arrangement for scientific and 
      high-performance computing systems

I'd like to see as much of the above as possible included in Fedora
Extras and [given what little free time I have! :-)] I'm doing what I
can to try and get the necessary parts packaged, etc.  I have a modules
package in progress and will be glad to share my unfinished bits with
anyone who is interested.

Do you have any objections to the above or maybe suggestions for
improvements?

Ed

-- 
Edward H. Hill III, PhD
office:  MIT Dept. of EAPS;  Rm 54-1424;  77 Massachusetts Ave.
             Cambridge, MA 02139-4307
emails:  eh3 at mit.edu                ed at eh3.com
URLs:    http://web.mit.edu/eh3/    http://eh3.com/
phone:   617-253-0098
fax:     617-253-4464