Bind forwarders

[Kevin Darcy]

If you've read some of my other posts on the subject, you perhaps know that
I have a fairly dim view of using forwarding for performance. The main problem
is that you already start in the hole when you do that. What I mean is that
you're already adding a mandatory hop to *every* query that you can't satisfy
from your own cache and authoritative data. To offset that mandatory
performance hit, you need an *awfully* fast and/or big forwarder or forwarder
farm (as well as a fast connection to that forwarder or forwarder farm) with a
high cache hit ratio for the queries you're sending there. Otherwise you're
better off overall eliminating that hop and just fetching the data yourself.
I'm sure there are network architectures where forwarding makes sense from a
performance standpoint, but I expect they would be very much the exception
rather than the rule. Personally I prefer for each server to operate
autonomously as a caching-only server. Less inter-machine dependencies means
more manageability.

I will note, however, that one of my biggest gripes against forwarding -- its
non-robustness -- has been mostly mooted by the changes in 8.2.3 (and
presumably also in BIND 9) which allow a sophisticated selection of forwarders
instead of the traditional sequential selection (which used to add a
significant timeout to every forwarded query when the first forwarder happened
to be down or unavailable).

Of course, if you're using forwarding *exclusively* for performance reasons,
make sure you define your forwarding mode as "forward first" so that you'll
still be able to resolve names even if all of the forwarders are down and/or
unavailable. But you probably already knew that.


- Kevin


Forrest Aldrich wrote:

> There doesn't seem to be much documentation on the use of Forwarders.
>
> For example, in our situation, we have PoPs all over the US which are
> running their own full secondary DNS.
>
> We've are regularly subjected to Visual Networks evaluations, etc., and we
> thought that our DNS performance would improve if we were to utilize
> forwarders within our own network cloud.  So, we have primary servers on
> east and west coasts that are rigged with lots of RAM, etc.
>
> I wanted to get more of an idea about the forwarder algorithm.  I know it's
> supposed to be "better" in Bind-9.  And the ideal application of a
> forwarder setup.
>
> For example, I thought that if we had different PoPs forward to other busy
> PoPs before our Primary DNS, that it might perform better.  But that
> becomes difficult to manage.
>
> Any pointers would be appreciated.
>
> _F