httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dirk-Willem van Gulik <>
Subject Re: Interesting Problem Report
Date Tue, 28 Sep 1999 17:04:22 GMT

On Tue, 28 Sep 1999, Rasmus Lerdorf wrote:

> Anybody have a clue on this one?

Hmm, any special modules ? What is MaxRequestPerChild set to? 

Though it _should_ not leak with this setup, it well might. 

Plus specicially with 2.2.5-22 I've seen the TCP stack crapping;
both on closing of connections and when a new one comes in some linear?
search getting slowly synced up when the load is more or less regular and
then getting inefficient at finding the holes and arranging the queue. So
then 1 out of X times the connection thus fails. Because it is out of the

Unfortunetely my last customer only wanted me to fix it, so we moved to
freebsd :-(, and we never had the time to really investigate. It was not
seen again on later installations.

> From: Bret Hester <>
> Newsgroups: comp.infosystems.www.servers.unix
> Subject: Apache performance slowdown problem
> Hello, all
> We have a performance problem we cant seem to solve
> with Apache 1.3.9 under Linux kernel 2.2.5-22. on intel
> hardware.
> Our Apache server is handling around 50+ requests a
> second..... all goes well for some hours 2-5 then Apache
> slows down to a crawl -+10 less requests a second.
> Our max servers setting of 150 is never reached.
> There are only ever around 70 apache processes
> running at one time.
> There is a increase in our failed connection attempts viewed
> with netstat during this period of slowdown
> Apache cant deal with the requests fast enuf and I assume
> the ques are over flowing...
> Is it possible that Apache is coaking on threads?
> When the children die and respawn the server comes back
> to speed, but for 5-10min there is a critical slowdown.
> Also an intresting note is for every 10,000 requests a
> child SEGV's, is this normal behavour for Apache???
> We have pushed up our listen queue length in the kernel
> include/linux/net/socket.h from 128 to 1024, this made
> no difference to the problem.
> We are wondering if perhaps this could be a socket
> serialization problem....   But we are running a single
> socket server....  what options might we try  here?
> Someone out there must have encountered this problem
> before, if so or just have ideas.... please help
> Thanks,
> Bret

View raw message