Am Mittwoch, 28. Dezember 2005 16:23 schrieb Sergio Freire:[...]

I don't have a serious test, but an observation I made that or may not 
be senseful in real life; don't know the exact version, but would 
surely be somthing like 1.8.x, maybe even 1.9.1.

What I did was a little stress-test to see how well lighttpd performs 
compared to apache; did this with apaches "ab", pulling only a static 
image; and then tried the same with a pound instance in front of 
either. pound and the respective backend (only one, so no real 
balancing involved) were running on the the machine, as well as ab 
itself (not very realistic scenario, I know).

What I saw: you can hammer lighttp alone with a huge concurrency, I 
think I went to above 10,000, and the cpu usage would still be very 
low. if you put a pound in front, the cpu-usage of pound goes up very 
quickly, using almost all resources, and the overall request/second 
ratio drops drastically.


Cheers, Sascha
[...]

Am Mittwoch, 28. Dezember 2005 16:23 schrieb Sergio Freire:[...]

another small note that might be of interest for your tests: you might 
get in trouble if your backends perform differently, and your 
priorities are not chosen accordingly. for example, a very slow backend 
may drop the overall performance of a cluster of equally prioritized 
backends (there was a nice explanation from robert at 
http://www.apsis.ch/pound/pound_list/archive/2004/2004-05/1084281441000).


Cheers, Sascha
[...]

On Fri, 2005-12-30 at 19:13 +0100, Michael Redinger wrote:[...]

This should be fixed in 2.0b5. Let me know if it isn't.
[...]

This would introduce enormous complications in the code for little
benefit. Please note that the cookie would be seen first (as it is
returned from the server), a session would be created and then another
session would appear due to the URL. At this point you have two
(possibly conflicting) sessions - not an enjoyable situation.

Why don't you use two separate services, each with its own session
mechanism, and a URL (or header) condition to separate them?[...]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1



Robert Segall wrote:[...][...][...]


Yes, the problem still exists for me in 2.0b5.

Michael


- --
Michael Redinger
Zentraler Informatikdienst (Central IT Services)
Universitaet Innsbruck
Technikerstrasse 13                    Tel.: ++43 512 507 2335
6020 Innsbruck                         Fax.: ++43 512 507 2944
Austria                                Mail: Michael.Redinger(at)uibk.ac.at
BB98 D2FE 0F2C 2658 3780  3CB1 0FD7 A9D9 65C2 C11D
http://homepage.uibk.ac.at/~c102mr/mred-pubkey.asc
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org

iD8DBQFDu8zVD9ep2WXCwR0RAuDyAKC/icE9EL5ihQatXdRrWd2pGd69CQCgs+IZ
rpiodkAjXxv8ExK6237oGVM=
=0+bc
-----END PGP SIGNATURE-----

On Wed, 2006-01-04 at 14:25 +0100, Michael Redinger wrote:[...]

OK - this is hopefully fixed in 2.0b6. Let me know how it looks.[...]