-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Mon, 1 Sep 2003, Shane Hollis wrote:

> The DNs is a semi heiracrchical system from my understanding. The master DNS ( 
> to use a phrase loosely) are based in the states. My provider in the states 
> is on the main backbone. Therefore changes get sent out much more rapidly and 
> from a centrally located place, not from a little backwater like NZ.
> 
> If I change a DNS here in NZ, it is like injecting an agent into tip of a 
> whales tail. It takes along time to propogate through the system. If I inject 
> into the heart it is sent out a lot more quickly and spreads like a ripple 
> from the centre of the pond, not from the edge. (some very mixed similes 
> here).

That's not how the DNS works. 

The DNS is organised as a heirachy purely for the purpose of splitting out
where answers come from into seperate entities, so that there is no single
central set of servers who are responsible for "all" answers. 

The root servers only know about domain names which are at the global
level, and that knowledge is limited to where to find answers. The servers
they point to have knowledge about the deeper parts of the DNS, and in
turn may point to other servers which inturn know more about even deeper
parts.

So when you look up, say, "hairy.geek.nz", there are a number of queries
which are sent out. The first is to find out who is authorative for .nz,
and on asking the root servers a list of nameservers will be
returned. Then, we ask one of those nameservers who is authorative for
geek.nz, and a list of nameservers will be returned. We then ask one of
those nameservers who is responsible for hairy.geek.nz and a list of
nameservers is returned. Finally, we now have a set of nameservers to
which we can ask questions about hairy.geek.nz.

I've simplified it, because the queries are not exactly that, but the net
effect is the same: We keep looking for an authorative server deeper and
deeper into the name.

Now, the problem with this system is it takes a while (not long, seconds,
but long enough) and that it involves a lot of queries and network traffic
as a result. So the answers given are _cached_, which means you get
answers quicker, and there's less traffic which means the 'net will
actually function. :)

How long it is cached for is determined by the TTL on the answer itself,
which is tunable per-answer. 

Thus, how long it takes for a change to propogate has nothing to do with
the depth in the tree, it depends on what parts of the answer are cached,
and how long they are cached for.

Since a name which has never been asked for before is never able to be
cached (beyond negative response caching, but that's usually not long),
the propogation time for a new entry at any level is basically zero. 

A name that has already been looked up will take as long as it's TTL to
"propogate", or more accurately, for the cached copies to expire and new
copies be asked for. This can be up to 48 hours, but it is entirely
tunable. It can be as low as 1 second if you wish, but I wouldn't suggest
doing that :)

The important thing to understand is that it DOES NOT MATTER where you are
in the network, the propogation time is entirely based on caching of the
answers, and nothing else. 

- -- 
David Zanetti           |  (__)             
#include <geek/unix.h>  |  ( oo    Mooooooo 
http://hairy.geek.nz/   |  /(_O ./         
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: Made with pgp4pine 1.75-6

iD8DBQE/Uo1PT21+qRy4P+QRAq2EAJwOk96Sk0xA1x3ys77axakA4hvSqQCgnyJO
cDMIYc1C/d10y316NTEb6/4=
=JhJA
-----END PGP SIGNATURE-----


Reply via email to