BIND 9 scaling
Phill Wood
phill.wood at gmail.com
Wed Feb 16 19:48:30 UTC 2005
On Wed, 16 Feb 2005 09:08:25 -0800 (PST), Rick Jones
<foo at bar.baz.invalid> wrote:
> Phill Wood <phill.wood at gmail.com> wrote:
> > I've just subbed to the list. We've been using BIND for some years
> > now, currently on versions 9.2.3 and 9.2.4.
>
> > We're master for a relatively small number of zones (~ 1500) and
> > provide a caching nameserver service as well. I've never encountered
> > any performance issues myself but my management are asking what the
> > limits might be.
>
> > We served 6,000,000 requests yesterday using one CPU of a dual-CPU
> > Sparc box (running with -n 1 due to threading bug in 9.2.3). Maximum
> > load average was 0.46.
>
> > Can I say, therefore, that we will be able to serve 12,000,000
> > requests easily and then performance may wane? And at what level of
> > hits would we start to suffer?
>
> You could say it, but I would not count on it being true unless and
> until you actually measured it or at least had a bit more supporting
> data.
>
> Load average is one thing.
>
> CPU utilization is another.
>
Yes, very true.
> Sometimes they relate, sometimes not.
>
> What was the CPU utilization of the one CPU running named when you
> served 6 million requests in a day? Particularly at peak?
>
Erm ... 96% idle.
> BTW, for "things networked" it has been my experience (via netperf)
> that one cannot entirely trust the output of mpstat - it misses time
> spent processing interrupts and when networking is involved, there are
> generally more than a few interrupts... Believe it if it says 100%
> utilized, believe it less and less as it reports lower CPU util -
> particularly if there is a non-trivial number of interrupts being
> reported.
.. thanks to everyone who has responded so far. Your answers have
certainly given me a good deal to think about and enabled me to give
at least a ball-park answer for now :-)
I'm hoping that I'm going to be able to get hold of some similar
hardware to my live system soon so I that can do some decent
performance tests.
Cheers
Phill
More information about the bind-users
mailing list