httpd-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jeff Trawick <>
Subject Re: [users@httpd] Apache HTTPD Sporadic High CPU
Date Tue, 10 Apr 2012 23:17:29 GMT
On Tue, Apr 10, 2012 at 5:51 PM, Bill Bawcombe
<> wrote:
> We are sporadically experiencing high CPU with our Apache instances.
> The CPU averages about 5% utilization normally and has run fine for over
> a year without an issue but over the past month we see spikes up to 100%
> utilization that never drop until the instance is bounced.  We can go up
> to a week before we see the issue but we've also seen it happen multiple
> times in a day.  The logs are not indicating any errors for further
> investigation and we only have this issue in our Production environment.
> Things we've tried:
> -We've searched numerous topics on Apache HTTPD and high CPU and with
> regards to tuning parameters but considering we haven't had an increase
> in usage and our average CPU utilization is normally low we are
> reluctant to make tuning changes without understanding what else we may
> impact.
> -We ran prstat and the one anomaly we've seen is the Involuntary Context
> Switching (ICX) seems to be high when the CPU spikes.  Normally ICS is
> in the single digits but when the CPU spikes we see thousands.  From
> observation it appears that the ICX value and CPU utilization increase
> and drop simultaneously.
> Information regarding our environment:
> - Apache HTTPD 2.2.20
> - MOD_PROXY as a reverse proxy
> - Solaris 10
> - 2 servers for high availability and have experienced the issue on both
> servers
> One other note is that we are using CA's SiteMinder (authentication and
> authorization) agent for Apache.  We've exhausted all of the support CA
> will give us regarding this issue and they have no other known issues
> from their customer base.

What kind of information did CA gather?  SiteMinder has its own
threads in the httpd processes, and of course SiteMinder can run on
the normal httpd threads.

One obvious thing to try is to get a handful of pstacks of a high CPU
process at 10 second intervals when the problem occurs and see if you
can find anything out of the ordinary there.  You can also point truss
to particular LWPs (as shown by pstack) for threads that look

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message