Beowulf start-up?

Robert G. Brown rgb at phy.duke.edu
Fri Feb 23 13:58:20 PST 2001


On Fri, 23 Feb 2001, ILGIN SEKER wrote:

> Hi to everybody. I'm new to this amazing list and have some questions:
>
> What are the "essential" hardware to build a beowulf? i.e. Can I get 10
> old 8085s (yes I have;), connect'em and make it?

8085's?  Not 8088's?

To make a cluster of sorts requires computers and some means of getting
programs and data between them.  In the earliest of days this might
have been an 8" floppy, a tape, or a serial line.  Such a cluster would
not be a beowulf.  If you really have one of the predecessors of the
venerable IBM PC a) I salute you; b) find good museums to donate them
to (keeping one, of course).  I still have an IBM PC squirrelled away
in a closet and plan one day to trot it out and mount a current
motherboard in it just for laughs.  Consider something like this.

Seriously you have to be able to run a real operating system on them to
build something that would be considered a beowulf -- I'd guess that you
MIGHT be able to do something with a 386, I'm certain you could do
something with 486's and up.  To build the simplest beowulf you'd need N
boxes with ethernet adapters and disks and memory sufficient to boot and
run at least a minimal linux kernel (including network driver) with room
left over for a network application.

However, with amazingly powerful nodes costing as little as $500 (or
less if you shop hard) there isn't a lot of incentive to build something
out of real antiques.  Even if you figured out some way of making it
work, you'd spend more on a year's electricity to run a collective 10
node network 1/10000th (or so) the speed of a single node that same
money would buy.  Currently I wouldn't mess with anything less than a
P6-class CPU unless it was strictly hobby stuff.

> And the same for software? All those Process ID 'globalers', file
> managers, control scripts etc.are confusing for me. What I need and what
> I've better have?

To start with, standard linux + PVM and/or MPI will probably do you.  Or
Scyld, if you want to expend what possibly would be more energy and
frustration on your early learning curve to extract a greater benefit on
the far side.  Another way to experience a steep but interesting start
up curve would be to use raw sockets in a parallel application.

> I want to code an example software to demonstrate beowulf performance to
> my instructor; what type of operations/usage are the good ones to show
> the stronger parts of beowulf?

Both PVM and MPI come with demo applications that you can build and use
to explore simple parallelism and parallel programming paradigms.  There
are versions of a Mandelbrot set exploration tool and the povray
rendering package for them as well that are appealing eye candy while
demonstrating one of the simplest ways to get decent work out of a
cluster.

You might look into reading over a number of resources linked to
www.phy.duke.edu/brahma.  There are some simple introductory papers and
presentations, a draft of a beowulf engineering book under development,
some tools (some of which themselves illustrate a kind of parallelism),
and links to lots and lots of other key resources, including the beowulf
HOWTO and FAQ, the beowulf underground site, the main beowulf site (of
course) and more.


   rgb

>
> That's it, thanks...
> Cagil SEKER
>
> "There's one thing stronger than all the armies in the world;
>  and that's an idea whose time has come." - Victor Hugo
>
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu







More information about the Beowulf mailing list