dns server missing .1% of the time!?

Nate Campi nate at wired.com
Thu Sep 6 07:44:03 UTC 2001


> > ----- Original Message -----
> > From: "intergate" <jason at vancetech.com>
> > Newsgroups: comp.protocols.dns.bind
> > To: <comp-protocols-dns-bind at moderators.isc.org>
> > Sent: Wednesday, September 05, 2001 7:35 PM
> > Subject: dns server missing .1% of the time!?
> >
> >
> > > I'm running a Redhat Linux 7 dns server that is hosting multiple sites,
> > > mvmeander.com and alienwizard.com From time to time ( I only notice once
> a
> > > week ) my DNS is just unavailable to resolve and domain requests and the
> > > client sees "Cannot find server or DNS error!".  I even got an email
> from
<snip>

I had a couple BIND nameservers that were serving authoritative info to
the internet and acting as "caching nameservers" for a bunch of hosts in
a datacenter. These two machines would stop responding for several
minutes out of every hour during peak load. What was weird was that it
was at one hour intervals.

I examined the hosts, and found that they were configured to store
'host-statistics' so you could dump stats and see who is querying your
nameserver. The problem with this is that it takes up lots of memory to
do this. I also dug into the docs and saw that BIND walks it's cache
every hour by default to clean out expired entries, and that if the 
machine is low on memory that it could cause it to page heavily during
this time. This sounded like my problem.

I commented out the 'host-statistics' lines in named.conf (default is
no) and things got better right away.

So whatever your cause, you might need more memory, and you're seeing
the same symptoms that I saw.
-- 
Nate Campi, UNIX Ops WiReD SF, Terra Lycos DNS, (415) 276-8678  


More information about the bind-users mailing list