Re: maximum/optimal # of SMP CPUs between 2.2 and 2.4

[Matthew Wilcox]

On Wed, Aug 09, 2000 at 03:13:33PM -0700, Larry McVoy wrote:
> Not to rain on anyone's parade, I've done my own testing of 2.4 and for
> some networking stuff it is literally twice as good as 2.2, but I find
> the statements below to be pretty misleading.
> 
> Aren't these quoted results from *static* web page tests?  And didn't 
> Linux use khttp?

specweb99 includes dynamic content.  TUX is an in-kernel web server (and
far much more according to ingo, but i'd like to see the code first).

> For us, in this forum, the point of benchmarks should be to expose where
> there is room for improvement, ideally compared to a theoretical limit
> (Linus' definition of "the best", one that I agree with), or compared to
> some other platform.  Looking at the results below, you'd think that Linux
> screams.  And it does, but on what I believe is an apples to oranges 
> comparison.

that's why I only compared linux results with linux results and nt
results with nt results.  i'd be interested to know if you think that
my assertion that `linux scales to 8 cpus better than nt scales to 4
cpus' is incorrect, based on the numbers below.  ignore the raw numbers.
the per-cpu degradation is what i'm interested in comparing.

> On Wed, Aug 09, 2000 at 02:55:58PM -0700, Dr. Kelsey Hudson wrote:
> > Actually, reading that it appears to me that Linux scales to 2 CPUs better
> > than NT scales to 4... Compare 2200 to 1598...

i think you're fundamentally misunderstanding the use of the word `scales'.
you're using it to mean `performs'.

-- 
Revolutions do not require corporate support.