CPU Usage

[Jeff Lightner]

We previously put our DNS servers in between external and internal
firewalls.   Initially I saw a very high CPU load because my acls in
named.conf were referring to the external facing NIC's IP/route etc...
rather than the internal facing NIC's.   On flipping that it got rid of
my high load.   NOTE: The default route is on the external and we have a
static route for the internal - that's why I had originally thought I
needed to use the external NIC's settings in acls.   If you have
multiple NICs and similar routing you might want to investigate this.

-----Original Message-----
From: bind-users-bounce at isc.org [mailto:bind-users-bounce at isc.org] On
Behalf Of Gushi
Sent: Friday, June 22, 2007 12:53 PM
To: comp-protocols-dns-bind at isc.org
Subject: CPU Usage

Hello all,

I am airing a previous issue.

After any amount of time from a day to a couple weeks, my monitoring
system will report that my BIND 9.4.1 server (built from ports) on
FreeBSD 6.2 is not responding.

Top reports that named is using 97 percent CPU, and about half the
time will not even respond to a standard kill.  Even when it IS
responding, it's at about 50 percent CPU.  What the heck is it doing,
and how do I find out?

I'm finding this in my logs:

clients-per-query increased to 26

but a google search for clients-per-query increased on the groups only
shows one bug, related to DLZ, which I'm not using.  And of course
since this is a new option, my printed docs (all the great o'reilly
books) are useless.  What does this mean, and is it a clue to the
problem?

These servers are masters for about 100 zones, and also recursive
nameservers for about 1000 machines.

I previously reported this problem in this thread:

http://groups.google.com/group/comp.protocols.dns.bind/browse_frm/thread
/e533629b0f0bbe2a/7564602871c06b77?lnk=gst&q=gushi&rnum=2#7564602871c06b
77