Possibly odd question

[Kevin Darcy]

John Oliver wrote:

> I had a client ask me today if he could have two separate nameservers
> for a domain, each pointing to a completely different IP address.  Let's
> say DNS A is 10.1.1.1 and DNS B is 172.16.1.1  DNS A resolves
> www.domain.com to 10.1.1.2 and DNS B resolves www.domain.com to
> 172.16.1.2  Besides the issues of synchronizing content between the two
> machines... is there a reason why this absolutely should not be done?
> His rationale is that if the connectivity for 10.1.1.0 goes down, then
> (after cached zone files from that server are purged) all future
> requests will fall back to the 172.16.1.1 serevr until the 10.1.1.0
> network is available again.  it sound sintriguing, but there's got to be
> a reason why nobody does this... :-)

Nobody does this? Lots of load-balancing "solutions" do this kind of thing.

The major downside is that you have to lower your TTL values a lot in order
to derive any benefit from it, and when you do that you risk
performance/abuse problems -- "abuse" in the sense that you're not only
making your own nameserver work harder; you're also making everyone else's
resolvers work harder trying to resolve the name, and thus committing a
"tragedy of the commons" transgression. If everyone did this, then caching
would basically be defeated and DNS would be way less efficient -- and thus
less useful -- than it is currently.

If all you're trying to do is provide network redundancy, you should do
this at lower layers. Usually folks resort to these kinds of DNS games to
provide load-balancing across server farms, but even in that case it's a
very imperfect solution.


- Kevin