On Tue, 5 Jun 2001, David Merrill wrote:
> On Tue, Jun 05, 2001 at 08:33:16AM +0100, James Sutherland wrote:
> > On Tue, 5 Jun 2001, Jamie Walker wrote:
> >
> > > On Mon, Jun 04, 2001 at 06:32:37PM -0400, David Merrill wrote:
> > >
> > > > Linuxers cried foul because system tuning wasn't done properly on the
> > > > Linux box.
> > >
> > > The other point was that the hardware spec and indeed the tests run were
> > > very atypical. There can't be many Web sites out there serving static-only
> > > pages over four 100 megabit ethernet cards.
> >
> > Good point - SPECweb includes a fairly large proportion of dynamic
> > content, to make it a bit more realistic. Tests these days are being done
> > on machines with 32Gb of RAM and 8 Gigabit Ethernet cards serving about
> > 66% static content from big RAID arrays, which IMHO is getting a bit
> > silly: are there *ANY* sites serving 8 Gbit/sec of content from a single
> > server?!?
>
> This, like most benchmarks, is valid as long as you carefully look at
> what it is measuring. It isn't *intended* to measure real-world
> throughput. It is *intended* to make the network speed irrelevant, so
> that the throughput of the machine *itself* is being measured. That
> was, I believe, the point of having excess bandwidth.
Basically, you're measuring an irrelevant statistic of an atypical system
under unrealistic circumstances: wonderful idea...
Next up: we measure the air resistance of SCSI adapters in strong magnetic
fields! :-)
> Of course, that also makes the results of the test irrelevant in
> real-world systems.
Yep. Of course, if Linux is beaten by an MS product here we'll never hear
the end of it anyway...
James.
_______________________________________________
issues mailing list
[EMAIL PROTECTED]
http://www.linux.org.uk/mailman/listinfo/issues
