Re: Documentation.
i think this would be very useful (an all in one guide). so far i have yet
to find a guide that explains it all in steps. Most guides only give
pieces.I am actually just starting with clustering and i have found it
quite hard to get info on setting up and using pvm or mpi. In fact the
only clstering program that doesn't takes days to figure out how to
install is mosix.
my .02
rob
On Thu, 31 May 2001, Adam C Powell IV wrote:
> This is a good point, we've had a lot of people ask "I'd like to do
> Beowulf, where do I start?" I wonder if a "getting started" or "HOWTO"
> document would be helpful...
>
> Here are some first steps (assuming you already know how to install Debian):
>
> * Install rsh-client and rsh-server, and edit /etc/hosts.equiv on
> each machine, adding each machine's name so you can rsh from
> everywhere to everywhere. Oh- and make sure those are in
> /etc/hosts with their IP addresses! Test with something like "rsh
> hostname ls". Take note of the security implications of running
> rsh (e.g. you might want to isolate the network where you're doing
> this). Also, I think ssh moves rsh out of the way, which is a
> pain...
> * For MPI, install mpich and/or lam everywhere. Edit
> /etc/mpich/machines.LINUX and/or /etc/lam/bhost.def (bhost.lam?)
> to add the machine names so mpich/lam knows what's in the
> cluster. With lam, "man lamboot" and act accordingly.
>
> I don't use PVM, so I can't suggest an entry there.
>
> The next step beyond this depends on what you're trying to do with the
> cluster. If that's "solving PDEs" (e.g. fluid dynamics, heat/mass
> transfer, phase field), then I'd suggest:
>
> * Install petsc-examples, "cp -a
> /usr/share/doc/petsc-examples/examples .; cd examples; make
> PETSC_DIR=/usr/lib/petsc PETSC_ARCH=`petscarch` BOPT=O
> testexamples", preferably in an NFS directory mounted on all of
> the machines. This will build and run all of the test and
> tutorial programs, running many of them in parallel. Edit the
> tutorials to suit your needs, and you're done. (Note that the
> finite difference stuff works well, but finite elements are still
> primitive with PETSc; hopefully ParMETIS will get into woody so we
> can do dynamic mesh repartitioning in parallel... :-)
>
> For monitoring:
>
> * Install jmon everywhere, edit ~/jmonrc to add all of the machines
> in your cluster, run jmon and watch the pretty performance bars.
>
> Other entries anyone? Diskless? FAI? Mosix? Molecular dynamics/first
> principles? Is a single document of this type really a useful thing?
>
> Zeen,
> --
>
> -Adam P.
>
> GPG fingerprint: D54D 1AEE B11C CE9B A02B C5DD 526F 01E8 564E E4B6
>
> Welcome to the best software in the world today cafe!
> <http://lyre.mit.edu/~powell/The_Best_Stuff_In_The_World_Today_Cafe.ogg>
>
>
>
> --
> To UNSUBSCRIBE, email to debian-beowulf-request@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
>
>
Reply to: