[ATrpms-users] atrpms.net blocking?

[Axel Thimm]

On Sun, May 07, 2006 at 04:47:47PM -0700, David Rees wrote:
> On 5/7/06, Axel Thimm <Axel.Thimm at atrpms.net> wrote:
> > I just say: Ouch
> >
> > # grep ^KeepAlive /etc/httpd/conf/*.conf
> > KeepAlive Off
> > KeepAliveTimeout 15
> >
> > So the fault was on the server ...
> >
> > My bad, I didn't check the settings and FC4 had this by default off.
> > I just turned it on, let's see.
> 
> The reason it's off by default is that turning it on will increase the
> number of httpd daemons required to sustain the same client load.
> Turning on keep alive will now keep a httpd daemon busy doing nothing
> for up to 15 seconds if the client supports keep-alive, even if the
> client doesn't make any more requests. This most likely will make your
> problem worse.
> 
> The primary reason to turn on keep alive is that it does improve
> performance if the client is making a lot of small requests, but it is
> at the expense of the server having to keep track of each keep-alive
> connection. In the case of Apache, this means an extra process, but
> for squid which is select based no additional process is required
> which is why it is able to scale much better than Apache.

A large amount of users were hitting the server with yum/smart/apt
that do downloads lots of small files. So after a couple of hours
activated no site yet was blocked again, e.g. noone used more than 20
SYN/EST connection and noone used more than 40 in total. Previously I
would see 10-20 sites gathering within the minute.

The issue with yum/smart/apt was that they would fire multiple
requests which had to be served by different processes as the previous
process was still tearing down the socket.
-- 
Axel.Thimm at ATrpms.net
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.atrpms.net/pipermail/atrpms-users/attachments/20060508/a49761e4/attachment.bin