Eric Brown wrote:
> 
> Guillaume Pottier wrote:
> >
> > Ron Parker wrote:
> >
<...>

> 
> I think that PVM internally can (optionally?) use MPI.  The only

I don't think so, as spawning processes, as down in PVM and used in
pvmpov, is introduced in MPI 2.0 and to my knowledge very rarely
implemented, at least not in mpich. I think lam has it, but I don't like
the configuration possibilities. :-(
So I don't think so.

On the other hand I have a port of pvmpov to mpi on the very end of my
priority list :-)
Mostly because pvm is not that stable, at least on our SUNMP. I talked
to Harald Deischinger ( he did the port to 3.0 ) and he knows of some
strange behaviours of pvmpov (sometimes some blocks don't get rendered)
and he knows of no reason :-). So perhaps a port would be nice. But not
at all urgent in my world, so hang on hang on.

Axel

> requirement for PVM is that all of the processors can use the same file
> system (NFS works just fine).  With the current release of PVM you can
> only use unix operating systems (linux works).  The upcoming release
> will allow you to use PCs too.  The benefit of PVM is that it is
> hardware independent.  You compile the PVM library and the program (in
> this case PVMPov) for each platform.  Thus you can mix SGIs with Sparc
> Stations.  If you have an 8 processor computer you can spawn 8 pvm
> processes -- all which end up on the same physical computer but
> different chips.
> 
<...>

> -Eric Brown

Post a reply to this message