[vox-tech] beowulf cluster

[Peter Jay Salzman]

begin Ryan Detert <rcdetert@ucdavis.edu> 
> 
> I am looking for a good howto or a really clear book on setting up a 
> beowulf cluster.

there's a beowulf howto, but it looks abandoned.  the copy i have was
last updated in 1998.   there's a book on beowulf clusters in the lugod
library, and it was good at the basics, but not very detailed when it
comes to actually setting one up.

i found the best references were webpages of people who have set them up
in the past.  i recall some pages at santa barbara being excellent.
note that there's really not one way of setting up a beowulf cluster.
the term is more like an umbrella term for a bunch of systems that run
batch processing, scheduling software and uses some kind of message
passing library.   literally, there are as many different ways of
setting up clusters as there are people who set clusters up.

> I have 3 computers and I am wondering first off if it 
> would be easier to use NFS or having each node have a completely 
> functional OS.

i've found NFS to be kind of a trivial thing to set up.  the cluster in
physics has separate OS's, but i think that might be due to laziness.
:)   i'm think bill broadley from the math dept mentioned his nodes are
diskless or nearly diskless.

if you go the separate OS route, NIS (yp) seems like it would be
perfect.  however, i never did get NIS to work.  tried for 2 whole days
and came up with nothing, so i finally gave up on it.  :)

don't know how much you know about beowulf, but do you have a particular
application in mind?  they don't parallelize stuff by themselves; you
either need to run other people's programs linked to something like MPI
or brew your own.   not that it's hard, but it can take a lot of thought
to determine the best way of parallelizing a particular problem.

pete