You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@httpd.apache.org by Mike Abbott <mj...@trudge.engr.sgi.com> on 2000/09/07 00:02:55 UTC
AAP vs. Zeus (was Re: AAP patch for 2.0a6 available)
Jim Jagielski wrote:
> It would be interesting to see how it compares to your Zeus
> implementation ;)
Yes, it was interesting. Thanks for the suggestion.
On a dual-PentiumIII Linux 2.4 system we found that Apache/2.0 with the
state-threaded MPM runs at about 2/3 the speed of Zeus 3.3. With the
dexter MPM it runs at about 1/2 Zeus. For reference, Apache/1.3.12
fully accelerated with the AAP's patches runs faster than Zeus and
spends 1/2 as much time as Zeus and 1/4 as much time as Apache/2.0 in
user mode. Linux 2.2 exhibits negative scalability (two processors are
slower than one); 2.4 is much better. Here's the Linux 2.4 data
normalized to Apache/1.3.12:
software speed user / system time
-------- ----- ------------------
Apache/1.3.12 + AAP 1.00 6% / 94%
Zeus 3.3 0.92 13% / 87%
Apache/2.0a4 + STM 0.68 25% / 75%
Apache/2.0a4 + dexter 0.54 25% / 75%
There's plenty of performance work left in Apache/2.0.
--
Michael J. Abbott mja@sgi.com http://reality.sgi.com/mja/
Re: AAP vs. Zeus (was Re: AAP patch for 2.0a6 available)
Posted by au...@coremetrics.com.
What kind of testing did you do to arrive at these figures?
Austin
>>>>>>>>>>>>>>>>>> Original Message <<<<<<<<<<<<<<<<<<
On 9/6/00, 5:14:05 PM, Jeffrey "W." Baker <jw...@acm.org> wrote regarding
Re: AAP vs. Zeus (was Re: AAP patch for 2.0a6 available):
> On Wed, 6 Sep 2000, Mike Abbott wrote:
> > On a dual-PentiumIII Linux 2.4 system we found that Apache/2.0 with the
> > state-threaded MPM runs at about 2/3 the speed of Zeus 3.3. With the
> > dexter MPM it runs at about 1/2 Zeus. For reference, Apache/1.3.12
> > fully accelerated with the AAP's patches runs faster than Zeus and
> > spends 1/2 as much time as Zeus and 1/4 as much time as Apache/2.0 in
> > user mode. Linux 2.2 exhibits negative scalability (two processors are
> > slower than one); 2.4 is much better. Here's the Linux 2.4 data
> > normalized to Apache/1.3.12:
> >
> > software speed user / system time
> > -------- ----- ------------------
> > Apache/1.3.12 + AAP 1.00 6% / 94%
> > Zeus 3.3 0.92 13% / 87%
> > Apache/2.0a4 + STM 0.68 25% / 75%
> > Apache/2.0a4 + dexter 0.54 25% / 75%
> Well, that stinks. From what workload were these figures derived? Since
> you mention negative CPU scalability in Linux 2.2, I would assume that
you
> are spraying static content over a fast wire.
> -jwb
Re: AAP vs. Zeus (was Re: AAP patch for 2.0a6 available)
Posted by rb...@covalent.net.
> > On a dual-PentiumIII Linux 2.4 system we found that Apache/2.0 with the
> > state-threaded MPM runs at about 2/3 the speed of Zeus 3.3. With the
> > dexter MPM it runs at about 1/2 Zeus. For reference, Apache/1.3.12
> > fully accelerated with the AAP's patches runs faster than Zeus and
> > spends 1/2 as much time as Zeus and 1/4 as much time as Apache/2.0 in
> > user mode. Linux 2.2 exhibits negative scalability (two processors are
> > slower than one); 2.4 is much better. Here's the Linux 2.4 data
> > normalized to Apache/1.3.12:
> >
> > software speed user / system time
> > -------- ----- ------------------
> > Apache/1.3.12 + AAP 1.00 6% / 94%
> > Zeus 3.3 0.92 13% / 87%
> > Apache/2.0a4 + STM 0.68 25% / 75%
> > Apache/2.0a4 + dexter 0.54 25% / 75%
>
> Well, that stinks. From what workload were these figures derived? Since
> you mention negative CPU scalability in Linux 2.2, I would assume that you
> are spraying static content over a fast wire.
Actually, it doesn't stink. Consider the fact that Apache 1.3 was tuned
by Dean for good performance, and nobody has really sat down with 2.0 to
do performance tuning yet. We need a stable solid server before we can
worry about performance. There are also many places that this performance
stuff can be hiding. I would suspect that our performance will improve
dramatically if we just switch the default poll code in APR to use select
instead of poll. Don't worry about performance yet, let's get 2.0 working
solidly first.
Ryan
_______________________________________________________________________________
Ryan Bloom rbb@apache.org
406 29th St.
San Francisco, CA 94131
-------------------------------------------------------------------------------
Re: AAP vs. Zeus (was Re: AAP patch for 2.0a6 available)
Posted by "Jeffrey W. Baker" <jw...@acm.org>.
On Wed, 6 Sep 2000, Mike Abbott wrote:
> On a dual-PentiumIII Linux 2.4 system we found that Apache/2.0 with the
> state-threaded MPM runs at about 2/3 the speed of Zeus 3.3. With the
> dexter MPM it runs at about 1/2 Zeus. For reference, Apache/1.3.12
> fully accelerated with the AAP's patches runs faster than Zeus and
> spends 1/2 as much time as Zeus and 1/4 as much time as Apache/2.0 in
> user mode. Linux 2.2 exhibits negative scalability (two processors are
> slower than one); 2.4 is much better. Here's the Linux 2.4 data
> normalized to Apache/1.3.12:
>
> software speed user / system time
> -------- ----- ------------------
> Apache/1.3.12 + AAP 1.00 6% / 94%
> Zeus 3.3 0.92 13% / 87%
> Apache/2.0a4 + STM 0.68 25% / 75%
> Apache/2.0a4 + dexter 0.54 25% / 75%
Well, that stinks. From what workload were these figures derived? Since
you mention negative CPU scalability in Linux 2.2, I would assume that you
are spraying static content over a fast wire.
-jwb