On Fri, 2006-06-30 at 08:24 -0700, James Dorsey wrote:
> Just a quick general query. I have a LAM-MPI cluster made of "spare
> parts" that's spent the last 8 months impressing the hell out of
> everyone in the lab and generally exceeding expectations. But my needs
> have changed once again. I need to set up a BOINC server. To cut a long
> story short, for network administrative reasons, the head node of the
> cluster is the prime candidate to act as the BOINC / web server.
>
> Unfortunately, I'll need to install Linux on the head node to make this
> work. I've tried and tried to compile the BOINC server on FreeBSD, and
> I'm giving up. Life is too short.
>
> So, the question. Does anyone know any reason why I shouldn't be able
> to run LAM with Linux (probably FC5) on the head node and FreeBSD on
> the slaves? I'm not worried about setting it up, - I installed and
> configured the FreeBSD cluster manually from source. But if there's
> some obvious reason why it's not going to work, I'd REALLY appreciate
> someone piping up and telling me.
>
> For information, the cluster is based on unwanted i386 machines using
> an NFS share for "/home", and running an old beta version of LAM-MPI.
>
> As far as I can tell, everything required for a working cluster is
> either a standard *NIX feature, or it's a part of LAM-MPI. So it
> shouldn't be a problem. In theory. Right?
This should work without any problems. The only thing that can be a
problem is that you can't run FreeBSD binaries on Linux and vice-versa.
One useful feature of LAM is that you can boot a node but not have it
used for scheduling MPI jobs. You can do this in the app schema with
the schedule=no option to the host. Something like:
headnode schedule=no
compute_node1
compute_node2
Then you can run lamboot / mpirun / etc from the head node and have
everything actually executed out on the compute nodes.
Brian
|
