You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@httpd.apache.org by rewt rewt <re...@linux-elite.org> on 2012/11/21 15:04:55 UTC

[users@httpd] Re: users Digest 21 Nov 2012 12:02:28 -0000 Issue 4476

Hi Tom,
Thank you very much for your feedback.
Does the actual setting you provided compliant with 32Gb RAM and 8vCPU ?

And another question regarding:
"no point having 1024 httpd slots multiplexing to 10 PHP
workers if 90% of your requests require PHP."

Sorry for my bad skills in english but i don t understand what you mean by
"no point" ?

Thank you for your time and your help it is much appreciated!

David


I don't know worker, we use event MPM. but it is somewhat similar in
> that you have a number of worker threads spread across a number of
> processes. We use these settings on our reasonably high traffic
> reverse proxy:
>
> # Each httpd child has 64 worker threads
> ThreadsPerChild 64
>
> # Start 8 children immediately for 512 workers
> StartServers 8
>
> # Have at most 16 servers
> ServerLimit 16
>
> # Start another child if we have less than 128 spare worker threads
> MinSpareThreads 128
>
> # Kill a child if we have more than 768 worker threads spare
> MaxSpareThreads 768
>
>
> With a threaded MPM, MaxClients is ServerLimit * ThreadsPerChild, so
> this allows for 1024 max clients. Each httpd process has a maximum
> SIZE of ~174M, and a RSS of around 75MB.
>
> Depending upon how much memory you need to allocate to the PHP app,
> and the proportion of hits that will require PHP you may need to scale
> up/down - no point having 1024 httpd slots multiplexing to 10 PHP
> workers if 90% of your requests require PHP.
>
> Cheers
>
> Tom
>
>
> Is it HP-UX or mingw? Try disabling building of htpasswd tool:
> cd support
> touch htpasswd
> cd ..
> gmake / make
>
>
>

Re: [users@httpd] Re: users Digest 21 Nov 2012 12:02:28 -0000 Issue 4476

Posted by Tom Evans <te...@googlemail.com>.
On Wed, Nov 21, 2012 at 2:04 PM, rewt rewt <re...@linux-elite.org> wrote:
> Hi Tom,
> Thank you very much for your feedback.
> Does the actual setting you provided compliant with 32Gb RAM and 8vCPU ?

The server has much less RAM than that - only 4 GB. All this server
does is read requests from the net, proxy them to another server, and
return the responses back to the net, and with that configuration, we
use less than 1GB of RAM, fully loaded.

>
> And another question regarding:
> "no point having 1024 httpd slots multiplexing to 10 PHP
> workers if 90% of your requests require PHP."
>
> Sorry for my bad skills in english but i don t understand what you mean by
> "no point" ?
>

There are two parts to any web app - the web server serving files, and
the app itself serving dynamic content. This isn't apparent if you use
mod_php, since the application is embedded inside the web server, and
so each process created to serve extra users also creates another
instance of the app.

This is more obvious if you use fastcgi, since the app runs outside
the web server, and you can easily see how many app processes you can
run. If you can only run httpd and 15 app processes on your box, then
there is no point making httpd scale to 1000 clients, since 1000
clients could not be served in time by only 15 app processes.

By using worker or event you can easily configure httpd to serve 1000+
simultaneous requests, but this is useless if your app can only serve
10 people at once.

Cheers

Tom

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org