Updated hypermail archives today

[Scott Dybiec <sd...@humanfactor.com>]

Just updated the hypermail archives for this list at:

http://www.humanfactor.com/cgi-bin/cgi-delegate/apache-ML/nh/

to include messages through yesterday.

Quick question: Will the Apache Group be supplying a specially modified
version of Apache to the MindCraft performance testers to solve the
"thundering herd" problem?

$cott

Re: Updated hypermail archives today

[Dean Gaudet <dg...@arctic.org>]

On Fri, 14 May 1999, Dan Kegel wrote:

> One thing the Apache folks should do, though, is make sure
> the benchmark uses mod_mmap_static, if they think it will help.
> Nobody seems to use this module (well, one guy thought he was
> using it, but he forgot to configure it right).

None of the other servers get the advantage of having their entire
filesystem statically mapped from the get go, why should apache?

Just like we're bitching about ZD's recent bogus comparison of Apache/CGI
to IIS/ISAPI, what you're suggesting is equally bogus.

Face it:  Apache/1.x is not the worlds fastest webserver.  If you're a
linuxhead and just want to see a fast linux/www benchmark then use zeus. 
If you want to see Apache improved, then you're on the right mailing list. 
But 1.x will never be able to stand head to head on a static content
benchmark.  There are other reasons to choose apache, and the problem is
magazines choose to ignore those reasons.

> (I'm assuming that the new hybrid server is nowhere near ready
> for benchmarking.)

apache-apr as it stands could be benchmarked, nobody would complain, it's
a useful educational experience.  But it's not something a magazine would
print. 

Dean

Re: Updated hypermail archives today

[Dan Kegel <da...@alumni.caltech.edu>]

Scott Dybiec wrote:
> Just updated the hypermail archives for this list at:
> http://www.humanfactor.com/cgi-bin/cgi-delegate/apache-ML/nh/
> to include messages through yesterday.
Thanks.  Any chance that could be automated?
 
> Quick question: Will the Apache Group be supplying a specially modified
> version of Apache to the MindCraft performance testers to solve the
> "thundering herd" problem?

The linux-kernel people are hard at work on wake-one semantics for
accept and friends.  Ingo says some of this is in 2.3.1-pre already.
See http://www.kegel.com/mindcraft_redux.html for updates.

One thing the Apache folks should do, though, is make sure
the benchmark uses mod_mmap_static, if they think it will help.
Nobody seems to use this module (well, one guy thought he was
using it, but he forgot to configure it right).

(I'm assuming that the new hybrid server is nowhere near ready
for benchmarking.)

- Dan

Re: Updated hypermail archives today

[Dean Gaudet <dg...@arctic.org>]

On Thu, 13 May 1999, Scott Dybiec wrote:

> Quick question: Will the Apache Group be supplying a specially modified
> version of Apache to the MindCraft performance testers to solve the
> "thundering herd" problem?

Thundering herd is a kernel problem, bug kernel people.

Dean