[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: lenny rant



Thomas:

It would seem that mpm-itk can't prefork, since it doesn't know what
requests it will serve and thus doesn't know what user to switch to.
If the servers did pre-fork, they would have to remain running as root
until they get a request, so that they know what uid to downgrade to.

One idea would be to fork children with a bunch of threads, each under
a different uid/gid, then pass off requests to those children. This is
essentially what mpm-prefork used to do. However, that module is no
longer maintained and I think was removed from apache2.2 (it existed
in apache2.0).

There is a big warning on the doc page for that module too saying that
it doesn't work.

Hope this helps clarify some things.

Cheers,

Jonathan

On Thu, Apr 16, 2009 at 11:54 AM, Thomas Goirand <thomas@goirand.fr> wrote:
> Adrian Minta wrote:
>> Michael Moritz wrote:
>>> On Wednesday 15 April 2009 02:50:30 Craig Sanders wrote:
>>> > BTW, performance is roughly equivalent to apache2-mpm-prefork. AFAIK,
>>> > php still doesn't work properly under the threading mpms (not that
>>> > there's that much performance difference on linux between threaded and
>>> > pre-forking apaches - fork on linux is very lightweight).
>>>
>>>
>>> We did some benchmarking of apache2-mpm-prefork vs apache2-mpm-itk.
>>> Our results back then were roughly that apache2-mpm-prefork with php
>>> is twice as fast as the itk and when serving static content about 20
>>> times faster -- see for instance
>>>
>>>
>>> http://chtekk.longitekk.com/index.php?/archives/22-Speeding-up-MPM-itk.html
>>>
>>>
>>>
>>> mm
>> In real world the situations is a bit better. The web browser doesn't
>> close and reopen the connection for every element on the page. Instead
>> uses the same connection for all the other requests to the webserver. So
>> after the initial fork the rest of the page and the following pages are
>> served from the already forked child.
>
> Hold on you all here... I hear so many times "fork fork fork". But a
> correctly configured Apache should NOT fork that much, and have
> processes already there. Is it that itk doesn't "pre-fork" at all?
>
> Thomas
>
>
> --
> To UNSUBSCRIBE, email to debian-isp-REQUEST@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
>
>


Reply to: