[Beowulf] RE: Compare and contrast MPI implementations

[David Mathog]

"Brian D. Ropers-Huilman" <bropers at cct.lsu.edu> wrote:

> David Mathog said the following on 2005-12-15 12:53:
> > Thanks for the feedback so far.  
> > 
> > Now how about MPI coexistance?

> One simple answer is to have your job scripts establish your environment
> variables for you via exports (or setenv if you use those other shells).

Um, I don't see how in general.  

For instance in lam-mpi to set up the parallel environment one
issues a "lamboot" command that in turn sends an rsh (or ssh)
to each of the compute nodes to start a "lam".  Unfortunately
those fail because the first command lamboot sends is "hboot",
and that isn't in the default path. If it let me run a
different command, for instance:

export PATH=/blah/lam_mpi/bin:$PATH ; hboot

all would be fine.  But I don't see any way to convince
lamboot to do that.  Hmm, I suppose if a lamboot
was already in the default path, and that one in turn executed the 
the command above it would at least let lamboot start?  Trying
that...  Nope. Because using rsh the only things in the path
are /usr/bin:/bin, it isn't going through the bash login to pick
up .bash_profile.  I don't really want to put anything like this
in either /usr/bin or /bin.

Any other thoughts?

Thanks,

David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech