[Beowulf] RE: Capitalization Rates - How often should you replace a cluster? (resent - 1st sending wasn't posted ).

Robert G. Brown rgb at phy.duke.edu
Thu Jan 22 05:26:15 PST 2009

On Wed, 21 Jan 2009, Lux, James P wrote:

> I think there's a B-school thesis in this topic. It sure would be
> interesting, especially if you had a decent modeling tool that would let you
> plug in cost of money, funding profiles, etc.

I agree.  Of the various columns and articles I've written, the ones on
infrastructure are among the most enduringly popular.  I was quite
recently contacted by an exec from a "large manufacturer of PCs" who
asked permission to use one of them in internal corporate planning

This is right in the sweet spot of model complexity.  The dimensionality
of the problem is order of ten -- not unmanageably high -- and there is
fairly solid data to support model parameters.

I am reminded of one of my own all-time favorite documents, let's see,
what was it called, something like:

  Configuration and Capacity Planning for Sun Servers

It looks like there may be an update of it called C&SP for Solaris
Servers out there somewhere.  Dunno about that, but the original was
fabulous -- I had (maybe still have if I look for it) a hard copy as
well as a PS image (postscript because this was back in the early 90's
or late 80's).

Didn't have an actual model, but has most of the parameters for one.
Still didn't focus enough on infrastructure requirements, though, or
optimum rollover replacement.  Back then servers were UBERexpensive,
though, so the rule was "use them until they fall apart".  We started
with a Sun 4/110 server in maybe 1985 or 1986, upgraded it to a Sun
4/310 (motherboard replacement, same chassis) in 1990-something) and ran
it until we got serious and bought an actual multiprocessor server in
1993 or 1994.  In fact we kept running it even then for a couple more
years but the new server ran NFS and provided all of the disk and much
of the management.

Now, of course Linux has forced the commoditization of servers so that
Sun cannot charge the absurd margins that they did back then (bad,
perhaps, but not AS bad).


> Jim

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