Re: Dual Ethernet Slowness Causing Lousy Benchmarks? (was: A very
Marc Mutz (Marc@Mutz.com)
Wed, 30 Jun 1999 00:52:15 +0200
Jordan Mendelson wrote:
>
>
> Now if this was a static test, then the only difference between the first test
> and the last test was the number of ethernet controllers used... and look at
> the performance difference:
>
> Linux 4KB static pages, 1 CPU, 1 Eth: 960 requests/second
> Linux 4KB static pages, 1 CPU, 2 Eth: 1100 requests/second
>
> NT 4KB static pages, 1 CPU, 1 Eth: 950 requests/second
> NT 4KB static pages, 1 CPU, 2 Eth: 2000 requests/second
>
yepp. In fact it was only a single page. They wanted to check if the
servers were capable of saturating (?correct english?) a 100Mbit line,
then later, if this was still true for 2 100Mbit lines.
Obviously NT has no problems 'saturating' up to four (see Mindcraft; or
even more?) NICs, when the service requested is trivial. But then Linux
can't and, as other people in other threads pointed out, that should be
looked at. Besides giving M$ a sure win in benchmarks, current and
forthcoming ones, it does not fit the target of improving scalability
(when not restricted to SMP, but including multiple-I/O).
Marc
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/