You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@httpd.apache.org by Steven Roussey <sr...@network54.com> on 2002/12/31 19:14:25 UTC

What is the source of this 'Thundering herd' redux?

Hi,

I have been trying to track down a nasty performance issue with apache
1.3 on linux 2.4.18 (RH7.2). I have narrowed the list down to two
possible culprits (apache or the kernel). I use PHP as well, but I don't
think that is responsible.

The load spikes that occur look like all the processes are waking up
(looking like the old 'thundering herd' problem). The odd thing is that
just prior, no processes are running and the idle % soars.

http://www.network54.com/spikes/vmstat.txt

It appears that nothing can accept an incoming connection (maybe one
process is supposed to, but does not) and so it then goes and wakes
everyone up to get it handled.

Any guess as to why (and where in the code) this happens?

The kernel people say it is not their problem, that it is an apache
issue.

http://www.network54.com/spikes/

RE: What is the source of this 'Thundering herd' redux?

Posted by Bill Stoddard <bi...@wstoddard.com>.

I assume you are using the lastest version of 1.3.  Try experimenting with the
AcceptMutex config directive and see if selecting different values changes the
behaviour you are seeing.

Bill

> Hi,
>
> I have been trying to track down a nasty performance issue with apache
> 1.3 on linux 2.4.18 (RH7.2). I have narrowed the list down to two
> possible culprits (apache or the kernel). I use PHP as well, but I don't
> think that is responsible.
>
> The load spikes that occur look like all the processes are waking up
> (looking like the old 'thundering herd' problem). The odd thing is that
> just prior, no processes are running and the idle % soars.
>
> http://www.network54.com/spikes/vmstat.txt
>
> It appears that nothing can accept an incoming connection (maybe one
> process is supposed to, but does not) and so it then goes and wakes
> everyone up to get it handled.
>
> Any guess as to why (and where in the code) this happens?
>
> The kernel people say it is not their problem, that it is an apache
> issue.
>
> http://www.network54.com/spikes/
>