Pretty sure we're hitting up against a concurrency problem. I can reproduce the issue on a machine that's not in production with very similar specs by running ab (apache bench) on three hefty client machines with concurrency set to 1,000 on each machine. Doing some tests serving static content, I changed my config from prefork to worker, and the situation improved quite a bit; however, the PHP modules our site uses aren't thread-safe, so that's not a practical choice for the moment. On Tue, Aug 31, 2010 at 5:55 PM, Rainer Jung <rainer.jung@xxxxxxxxxxx> wrote: > On 31.08.2010 23:38, Eric Covener wrote: >>> >>> Running `netstat -tapn | grep 80 | wc -l' shows 11299 >>> connections to port 80. We're setup to use prefork with the following >>> configuration. >>> >>> MaxClients 1000 >> >> Is the netstat output a typo? With MaxClients 1000 and the default >> ListenBackLog of ~500, you should have started refusing connections >> long ago. > > Likely most of them are TIME_WAIT. Those are not connected any more, but to > many of them make the TCP connection hash slow on many platforms. > > If the OPs output is like mine, he can get a better statistics via: > > netstat -tan | awk '/:80 / {print $NF}' | sort | uniq -c > > Regards, > > Rainer > > --------------------------------------------------------------------- > The official User-To-User support forum of the Apache HTTP Server Project. > See <URL:http://httpd.apache.org/userslist.html> for more info. > To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx > " from the digest: users-digest-unsubscribe@xxxxxxxxxxxxxxxx > For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx > > --------------------------------------------------------------------- The official User-To-User support forum of the Apache HTTP Server Project. See <URL:http://httpd.apache.org/userslist.html> for more info. To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx " from the digest: users-digest-unsubscribe@xxxxxxxxxxxxxxxx For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx