[ILUG] DNS question

[Niall O Broin]

On Friday 5 March 2004, kevin.dobey at trintech.com (Kevin Dobey) wrote:

>Which reminds me - do DNS CNAMES have a time to live or does it inherit the
>ttl of the name it's pointing too (looks like the latter to me). 

This was a timely question. I was just thinking of something last night, the
objective of which is an inexpensive form of redundancy for web sites on
colo'ed boxes. 

The idea is that you have two boxes, each of which is in a different colo
premises. You have a domainname, say www.company.com, which is CNAME'ed to
company.dynamicdnsprovider.com. In the normal course of events, you have
company.dynamicdnsprovider.com pointing to colobox1.company.com. If, for
whatever reason, colobox1.company.com goes off the air, you just use
dynamicdnsprovider.com's tools to update the pointing of
company.dynamicdnsprovider.com.

So, what's wrong with this scenario? Well, one thing is that I have wondered
how dynamic DNS providers work at all. I'm no DNS expert either, but neither
am I completely ignorant about DNS, and AFAICT can tell the only way dynamic
DNS services can work is if they use DNS servers which update immediately, and
use very short TTLs on their zones. 

However, I always understood that you shouldn't use very short TTLs except
when about to change a domain to a different IP, as such TTLs are not net
friendly. Also, some ISPs' caching servers have a tendency to ignore TTLs as
specified in zones and cache for just as long as they feel like caching
(Eircom was a particularly bad offender in this regard - I don't know if they
still are).

If what I am surmising about the operation of dynamic DNS providers is
correct, then wouldn't setting short TTL on company.com's DNS servers work
just as well as a dynamic DNS service (while of coure being equally net
hostile) ? Then in the event of a failure of colobox1, the DNS could just be
updated to point to colobox2.

I really hope I'm not missing something glaring here :-)


Niall