From: Matthew French (mfrench42 at domain yahoo.co.uk)
Date: Wed 29 Aug 2001 - 22:09:44 IST
On Wed, Aug 29, 2001 at 07:54:53PM +0100, Gavin McCullagh wrote:
> I've noted several other groups with similar workloads buying Xeon
> based machines. We've not done this as is appeared to us the
> advatages would be limited and hence the money would not be worth
> it. Under what circumstances is it recommended to get a Xeon?
The best approach is to get your hands a Xeon temporarily and run a
comparison between the two using your software. While it may be
tempting to theorise about the benefits of the Xeon design, it is
usually more practical to demonstrate that the Xeon is effective - or
ineffective as the case may be.
You could find that the Xeon as a 10x improvement in which case it
would almost be a no brainer, or you could find the Xeon runs just a
little bit faster, but my experience has always been that the theory
is great, it is the implementation that counts.
Use the "time" command and run a sample set of data several times on
the Xeon and on one of your machines. From this you can then identify
how much the faster machine will save you and make a claim on this
Also remember that it is often more efficient (and more fun) to
optimise the code than to rely on faster hardware. I remember one
instance where I reduced a series of FFT calculation from 40 minutes
to 10 seconds for the same piece of data. On another occasion I
reduced a database load from 1 hour to 1 minute. This took me roughly
two days effort for each, and the performance improvement only
required a change in algorithm and a rewrite of the C++ code.
That is a 60x improvement over the 3x improvement a faster processor
_might_ get you.
 You talk about recompiling, so I am assuming that the code is
available to you.
Do You Yahoo!?
Get your free at domain yahoo.com address at http://mail.yahoo.com
This archive was generated by hypermail 2.1.6 : Thu 06 Feb 2003 - 13:11:53 GMT