[buug] Clusters

[Michael Salmon]

On Wed, Sep 04, 2002 at 10:19:10AM -0700, Mark Hedges wrote:
> 
> On Tue, 3 Sep 2002, f.johan.beisser wrote:
> > beowulf used to be a pain to set up and design. the advantage, of course,
> > is that with a little programming savy, you could have an MPP machine for
> > much less than the cost of a Cray.
> 
> There's a Debian package called FAI (fully automatic
> installation) that the creator wrote primarily to set up a
> cluster of beowulf machines.  It should make things easier.
> 
> I was wondering myself, do beowulf nodes have to run on
> identitcal machines?  Can a beowulf run on disparate units?
> Would it just bottleneck processing speed to the slowest CPU?
> 
> Does anyone care to explain the difference between parallel
> computing and distributed computing?

    Distributed and parallel computing are spoken of nearly interchangably
because for most people they mean the same thing.
    Distributed computing is based on parallel computing and it basically is
parallel computing over a network with hetergenous components, while parallel 
computing began on multi-processor machines with homogenous components.
    In academics you'll see more about parallel computing then distributed
computing simply because to most researchers the distributed part of parallel
computing is just a new implementation of parallel computing. Computer science
classes will focus on parallel algorithms, like how to do 3d volume rendering
in parallel. Once the algorithm is designed, whether it runs on 2 cpus
on a smp machine or over two cpus over tcp/ip isnt very important. If the
distributed management framework is designed properly it simply wont matter.
    If you are interested in parallel computing, I recommend you learn a
parallel programming language like NESL. 
http://www-2.cs.cmu.edu/~scandal/nesl/tutorial2.html
    I've never used Beowulf. Since I dont have a degree in computer science
knowledge of parallel computing is basically worthless to me though.

> --mark--

ms