httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ian Holsman <>
Subject Re: 2.0.29 high load avg
Date Wed, 09 Jan 2002 17:18:51 GMT
I have put a recent test of Apache 2.0.29 up on webperf.

this test simulates a increasing Transaction Per Second rate over time
for a random URL on a NFS docroot.
The file itself has >10< included files.

The load is made up of
	20% 56K modem users,
	40% DSL Users,
	40% T1 Users.

The thing to note is we have keepalives turned off.
(The pause in the middle is due to the load generator screwing up
not the webserver)

The webserver is a Solaris 8 machine with 2G memory and 8 200MHZ processors

If any of this is unclear I'm put more detail in.


Greg Ames wrote:

> Justin Erenkrantz wrote:
>>On Tue, Jan 08, 2002 at 10:18:18PM -0500, Greg Ames wrote:
>>>as I mentioned in another post, 2.0.29 also seems to have the high load average
>>>problem.  The run queue size numbers displayed by vmstat weren't quite as
>>>extreme as what I saw yesterday pm, which may be because there's less traffic
>>>the evening.  But then the load averages displayed by top gradually built up
>>>until they were much higher than what we see with 2_0_28.
>>Well, I can reproduce it locally just by asking for the manuals
>>with flood:
>>jerenkrantz@walla% uptime
>> 8:05PM  up 11:51, 3 users, load averages: 22.64, 9.35, 3.76
> COOL!!  
>>Looking further into it now.  Of interest in the top output is that
>>a lot of process are "Mutex" wait.  
> uh-oh.  The live server on daedalus doesn't use accept mutexes.  The idle
> servers sit in accept() because we don't think FreeBSD has a thundering herd
> problem.
> do you see the "Mutex" wait in the STATE column?  and is walla FreeBSD
> w/prefork?  when this gets really bad on daedalus, top shows "RUN" in the state
> column for most of the httpd's briefly, then after the screen refreshes you see
> more normal waits - select or accept, for example.
> Greg

View raw message