DNS Scaling issues

[Jorge Amodio]

>> Also having the query being satisfied via TCP will potentially break
>> the use of ANYCAST as the
>> mechanism that enables to have replicated "mirror" root servers around
>> the world.
>
> Well Anycast servers do repsond to TCP DNS queries, but UDP simplifies
> things re: bandwidth and other issues.

The problem with Anycast is that the TCP state for the connection is maintained
only at the server with which the original connection was established, if by any
network event that produces a topology change, ie route change, following
TCP packets could be directed to some other server that has not the state
for that connection, then the query will fail and the client will have to retry.

> The main takeaway from the root scaling study report for me was that
> the proposed changes to the root are multiplicative in nature, and not
> simply additive, so do the big one first:

Agreed, it is not additive. Also one of the assumptions is that TLDs
will have average at least two NS records, but actually that number
will probably be higher.

> "A name indicates what we seek. An address indicates where it is. A
> route indicates how we get there."  Jon Postel

He had the ability to present the complexity of different pieces of the
Internet architecture with outstanding clarity and few words.

Another thing is that not many people realize that the DNS is not just
being used to get people to their favorite website, as I mentioned in
my previous message, one of the tools to deal with email spamming
relies heavely in DNS but that is something that happens under the
hood.

Regards
Jorge