httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Marc Slemko <ma...@znep.com>
Subject Re: ZD do it again!
Date Thu, 27 Feb 1997 23:45:44 GMT
On Thu, 27 Feb 1997, Brian Behlendorf wrote:

> On Thu, 27 Feb 1997, Marc Slemko wrote:
> > Wasn't in the store here when I checked.  I have no problem with reviewers
> > saying you have to have half a clue to make Apache work.  I have no
> > problem with them saying it does not include all sorts of gee-whiz gizmos.
> > However, I am concerned about the results they are getting performance
> > wise.  While there is certainly a hit due to Apache not being
> > multithreaded, in my tests against Zeus with the right server setup Apache
> > kept up with Zeus pretty well, and Zeus is probably on par or better than 
> > MS and NS's servers.  Largely because of the process per client model of
> > Apache, setting the server up correctly is more important than with a
> > threaded server because if it is done incorrectly it will be forking far
> > too often.  
> ...
> > The actual numbers are not on the web, and they may have more information
> > validating their methods in the article, but it looks quite bogus to me.
> 
> We need a web server olympics - an event where each vendor can spend $5000 on
> hardware to build their dream machine, tune it as much as they like, and then
> each submits a set of tests to be applied to all servers.  
> 
> Or if the point is that a server should be speedy without requiring a lot of
> clue as to how to configure the server and machine, perhaps give an "average
> Joe webmaster" the task of installing and configuring it.  

The thing is that for the sort of load these benchmarks are testing, you
would be a fool not to have someone who knows what they are doing with
whatever OS you use, so I really am not sure how big of an issue being
able to serve 500 hits/sec out of the box with no config file changes is.
Some magazines make it appear like that is important.

If you lower the hardware limit to $3000 you would be better off for
Apache.  <g>

> 
> This could be really cool, particularly as an in-person event.  Hmm!

Hey, if you want to fly me down from Canada... <g>  Didn't think so, guess
I'll have to cycle...

I am thinking about trying to contact PCMag, both to express my concerns
about their test and try to get the idea in their heads that it may be
wise to consider getting in touch with us for any future benchmarks to
review their config... they appear to admin to being clueless about Unix,
but that could just be one person and they may have had someone with a
clue do the Unix setup.  OTOH, they had trouble with the need to set the
undocumented User param, so... <sigh>

Most of the tests are on Stronghold, so I'm not sure if they did contact
them or not for performance tuning...


Mime
View raw message