[Beowulf] 96 Processors Under Your Desktop

[Robert G. Brown]

On Wed, 1 Sep 2004, Joachim Worringen wrote:

> John Hearns wrote:
> > On Tue, 2004-08-31 at 22:49, Jim Lux wrote:
> >>The real value (to my mind) is making a cluster a minimal-hassle item, the
> >>same way a desktop PC is perceived today....
> > 
> > You make a well-argues point.
> > Then again, that was the original Beowulf concept - use COTS components.
> 
> I can't see a correlation between "usage of COTS components" and 
> "minimal hassle". At least, not a correlation like "more COTS, less 
> hassle". Just the contrary, if anything.
> 
> A system which was designed for one purpose from top to bottom will 
> serve this purpose with minimal hassle (good design provided). A system 
> build from components for universal usage requires much more integration 
> efforts by people who have no influence on the design of the individual 
> components.

But there already ARE plenty of COTS cluster nodes, designed from top to
bottom to build compute clusters or server farms out of.  COTS doesn't
mean that the boxes have to be desktop units, just that they are
mass-marketed and readily available.  Nodes are currently available "off
the shelf" from many vendors as much as any web-based commodity is these
days of just-in-time assembly, testing, delivery.

This isn't quite as oxymoronic as it sounds -- cluster nodes from
Penguin, Appro, IBM, Dell are typically built out of commodity parts,
for all that they might have e.g. a custom designed case or a
server-class motherboard.  The fundamental design still uses off the
shelf chipsets and largely off the shelf components arranged in the
"usual" ways but effectively preconfigured at the hardware level for
computer cluster (or server farm) use.

To build a cluster we order "standard nodes", selecting a hardware
configuration from what the current crop of commodity choices permit.
It is delivered.  We rack 'em up.  We boot 'em and they PXE/kickstart
installs themselves, then yum-maintain themselves.  

Although this requires time and expertise to set up, it scales to an
arbitrary number of nodes with little additional work.  We use the nodes
until they break or age out to obsolescence and are retired, fixing the
hardware as long as it makes economic sense to do so.

Hard to get less hassle than this... although doing the same thing with
vanilla boxes in tower units with heavy duty shelving from a hardware
store isn't MUCH more difficult...

  rgb

> 
>   Joachim
> 
> -- 
> Joachim Worringen - NEC C&C research lab St.Augustin
> fon +49-2241-9252.20 - fax .99 - http://www.ccrl-nece.de
> _______________________________________________
> 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