New benchmark IIS vs Apache

[ri...@atm9.com.dtu.dk]

http://www.mindcraft.com/whitepapers/nts4rhlinux.html

Can someone give counterarguments for their claims?
Independently if serving static files is by any means a measure of the
performance of a web server, is their analysis correct?, is there anything
these people "forgot" to tune?

May it be appropriate that the Apache Group posts some kind of press release
or response to that?

Greetings

Daniel

Re: New benchmark IIS vs Apache

[Marc Slemko <ma...@znep.com>]

On Wed, 14 Apr 1999, Graham Leggett wrote:

> ridruejo@atm9.com.dtu.dk wrote:
> 
> > http://www.mindcraft.com/whitepapers/nts4rhlinux.html
> > 
> > Can someone give counterarguments for their claims?
> > Independently if serving static files is by any means a measure of the
> > performance of a web server, is their analysis correct?, is there anything
> > these people "forgot" to tune?
> > 
> > May it be appropriate that the Apache Group posts some kind of press release
> > or response to that?
> 
> I saw a number of comments on this already that disputed the claim that
> Mindcraft had "optimised both servers".
> 
> It was pointed out for example that the MinSpareServers setting was
> changed from it's default setting of 5 to a setting of 1, effectively
> starving Apache of available processes when loadspikes happened.

In this particular benchmark situation, there should be few if any such
load spikes.  MaxSpareServers is high enough to let the rampup period get
lots of servers running to deal with the load.  In fact, having the lowest
possible MinSpareServers is a _good_ thing for the load profile of this
particular benchmark because it minimizes wasted overhead.

> Perhaps a structured analysis of the configuration used, and whether
> this configuration was optimised (or de-optimised) would be a valuable
> addition to the argument.

What they posted of the Apache configuration was not obviously
de-optimized in any way that matters a significant amount.  Fixing it up
could give a 10, 20, maybe 30% increase in performance.  Apache would
still look almost as bad if you increased all its numbers by 30%.  I'm not
trying to shift any blame, but this is almost certainly an OS bug or
resource limitation/configuration issue.

Re: New benchmark IIS vs Apache

[Graham Leggett <gr...@dsn.ericsson.se>]

ridruejo@atm9.com.dtu.dk wrote:

> http://www.mindcraft.com/whitepapers/nts4rhlinux.html
> 
> Can someone give counterarguments for their claims?
> Independently if serving static files is by any means a measure of the
> performance of a web server, is their analysis correct?, is there anything
> these people "forgot" to tune?
> 
> May it be appropriate that the Apache Group posts some kind of press release
> or response to that?

I saw a number of comments on this already that disputed the claim that
Mindcraft had "optimised both servers".

It was pointed out for example that the MinSpareServers setting was
changed from it's default setting of 5 to a setting of 1, effectively
starving Apache of available processes when loadspikes happened.

Perhaps a structured analysis of the configuration used, and whether
this configuration was optimised (or de-optimised) would be a valuable
addition to the argument.

Regards,
Graham
--