Re: Application protocols and Address Translation

At 14:01 +0100 12/1/02, Patrik Fältström wrote:
>My number one argument for moving to IPv6 is that we have enough 
>addresses to have unique addresses on every device, so NAT will not 
>be needed in v6 only land.

Sadly, I wish I had more time to delve into this discussion. 
Recently we came up against this issue in the DNS realm (and DNS, 
perhaps surprisingly, is just an application in some architectural 
perspectives).

I am not settled on whether:

The end-to-end principle is an overriding concept that must be 
maintained in order to build network wide applications.  In the DNS 
situation, resolvers (clients) not being able to contact specific 
servers are causing some backwards compatibility problems.

The notion of an "inter-network" should imply that it is a network 
that connects networks.  In this case, applications should be built 
and/or restricted to ease the transition from one network (layer 3) 
to another.

Part of the issue is in the debate of a traditional smart-edge 
dumb-core network versus the traditional telephony assumption of a 
dumb-edge smart-core network, and all of the architectural 
implications that come with this.  With the growth of telephony 
discussions in the IETF, I suspect that the debate is becoming more 
relevant (again).

In the environment of DNS, I called the elements of DNS that sat on 
boundary devices (probably incorrectly labelled NAT ALGs) as 
"miscreant protocol elements" because they interfered with the spirit 
of the DNS protocol and application.  Either the elements were the 
problem, or it was a failure on the part of the DNS protocol to 
harden itself against such elements.

I don't know.  As I say, I'm unsettled and pressed for time on other 
topics right now.

-- 
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Edward Lewis                                          +1-703-227-9854
ARIN Research Engineer

Received on Monday, 2 December 2002 11:30:41 UTC