Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 22 Apr 1998 13:38:53 -0600 (MDT)
From:      Marc Slemko <marcs@znep.com>
To:        Don Wilde <don@partsnow.com>
Cc:        freebsd-advocacy@FreeBSD.ORG
Subject:   Re: [Fwd: Freeware]
Message-ID:  <Pine.BSF.3.95.980422132113.13481B-100000@alive.znep.com>
In-Reply-To: <353E34E3.308E0840@partsnow.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 22 Apr 1998, Don Wilde wrote:

> I just got this in response to my 'Challenge' posted in various places. Sounds
> like SPEC might be open to working with us. In looking over the solution

SPEC really can't do much.  You can't have the organization responsible
for unbiased creation and recording of benchmark results subsidizing one
vendor's tests.  They have donated a copy of SPECweb96 to an Apache Group
member for testing, though, but setting up a decent network to use it on
is another issue because it requires ugly OS configurations and the
license is location-limited.

The fact is that all the top results there take $$$ worth of hardware, not
just for the server but for some massive clients too.  That is hard for
most free software projects to deal with unless a member is lucky enough
to be in the right place at the right time with the right resources to
borrow.

> provided by Novell, it seems that what they did to achieve that number was to
> open up massive bandwidth in hardware.. I don't have any systems that have 5 PCI
> slots for 100Base-T ethernet cards, and I don't have any P-][ 300Mhz chips
> laying around. SCSI-3 we can do, fast-wide disks ditto. Do we have Fibre Channel
> or SSC boards available to us? Is ATM stable yet?

To kick your web benchmark results up high, you need to shove in
interfaces with big MTUs.

> 
> It'd be interesting to see how close we can get to that with less hardware,
> playing the same game that the other vendors do [RELEASE: FreeBSD/Apache
> Achieves 75% of Novell's SPECweb96 performance with 2/3 the Processor Speed!!!].
> Alternatively, presenting a real-world system would be more valuable to real
> users. Comments?
> 
>  http://www.specbench.org/osg/web96/results/res98q2/web96-980322-02570.html is
> the URL of the Novell results, and webmaster@specbench.org will get to them.

If you want to push FreeBSD based on web server benchmarks, you would be
better off doing so using Zeus (http://www.zeus.co.uk/) if you can get
Zeus to go for making a recent available version on FreeBSD, since there
isn't much that can keep up with Zeus, in general, although SGI is saying
their IRIX-ized NS Enterprise can top it at times. 

Take a look at http://www.zdnet.com/pcmag/features/webserver98/bench.html
and http://www.zdnet.com/pcmag/features/webserver98/bench2.html

This has Apache on Solaris and Linux, and it uses zdbench, but the results
are quite clear.  Unlike many previous published benchmarks, this is not a
result of a misconfigured Apache (well, not an obviously misconfigured
Apache anyway;  I walked them through the config for Apache on the Linux
box, and someone else was with them to verify that things were
workingright), and changing the OS to FreeBSD isn't going to improve
results by a factor of two or three times.  OTOH, the Linux box was
running with a single CPU enabled against dual CPUs on the other Intel
boxes, and I'm not sure offhand how the Intel hardware compare to what
Solaris was running on. 

Using 1.3 can probably give you a 10-40% boost, but Apache is not
optimized for benchmark testing like Enterprise and IIS are and trying to
show it kicking ass in that area is, for now, not overly productive. 

You could perhaps do better if you setup a complete dynamic content thing,
since a well written Apache module generating dynamic content can be
faster than serving static content. 

Once you factor in reliability and other aspects of performance under
load, you can well end up with a "faster" solution using Apache on FreeBSD
than something like IIS but that doesn't show up in benchmark numbers.


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-advocacy" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.3.95.980422132113.13481B-100000>