Robert G. Brown rgb at
Wed Jul 24 22:01:09 PDT 2002

On Wed, 24 Jul 2002, Jim Fraser wrote:

> With regard to scientific software pricing for clusters, what do you guys
> consider reasonable model for pricing?  Lets say the basic price for some
> given software is 50K for 1 CPU. What do you think is reasonable for
> increasing numbers of CPUs.
> NCPU's     COST $
> 1-4        50K
> 4-16
> 17-32
> 32-64
> 64-128
> 128-256
> 256+

Boy, a tough question to ask on a list devoted to COTS clusters loaded,
for the most part, with open source software whose cost scaling is more
like $50 (if that) for 1-256 CPUs.

Especially tough given that your quoted single CPU price is $50K.  For
$100K one can probably afford to hire an out of work programmer to write
you your very own open source version of nearly anything, so it is
probably a good idea to stay somewhere below that;-)

I'm afraid I'm with Martin on this one, especially where scientific
software is concerned.  I have no idea what you could be selling that is
worth $50K for a single CPU license (making the software cost close to
100X the hardware cost) in the scientific arena -- nuclear bomb
simulation programs?  Rational drug design software?  Something fancy
involving genetics?  Nor do you indicate how the application scales in a
"cluster" environment -- can it be run on 256 CPUs 255x as fast as on
one?  Can it be run on TWO CPUs any FASTER than on one (quite possibly
not if it is e.g. disk bound).

Still, presuming that it is a "real" cluster application that can scale
out to N CPUs (where N is a pretty large number, depending on cluster
architecture and design) that somebody somewhere is willing to pay as
much or more than they paid for the cluster itself to use, I'd use a
"site license" model for the cluster.  One price and you can use it on
as many CPUs as you like (or can scale to) in a given cluster.

However, I'd also hope that somebody in your market decides instead to
invest the time to write an e.g. gsl-based open source version of the
software -- you're providing plenty of incentive to do so.

Overall, I think that the $0-50 (per system) range is the ballpark of
"fair market price" for nearly any software, probably on a sliding scale
for clusters.  In that range, one can often convince folks to "buy" open
source software for the service, or for the convenience.  Even if it is
a proprietary closed-source application, one also provides them with
little incentive to engineer their own open source replacement.  When
one hits software prices with too many zeros this is often an attractive


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

More information about the Beowulf mailing list