W3C home > Mailing lists > Public > ietf-discuss@w3.org > December 2002

Re: NATmakes a network a host -- must every process have an IP address?

From: Edward Lewis <edlewis@arin.net>
Date: Fri, 6 Dec 2002 08:48:20 -0500
Message-Id: <a05111b00ba165994379a@[]>
To: Dave Crocker <dcrocker@brandenburg.com>, Edward Lewis <edlewis@arin.net>
Cc: discuss@apps.ietf.org

At 22:35 -0800 12/5/02, Dave Crocker wrote:
>Sounds like you are tripping across the end-to-end myth.  Whenever we design
>a protocol that believes there is a direct interaction between one end-point
>and the other, we later have to deal with the presence of intermediation.
>Caches.  Firewalls.  Whatever.
>Store-and-forward is not just for layer 3.

I'm not sure if I agree or disagree.  Yes, I'm tripping over an 
end-to-end assumption made in the protocol, but I don't get the 
'myth' part.

You describe the problem nearly dead on.  When using the existing 
cache mechanisms in DNS, they can become an obstacle if an old 
version is not compatible with a new feature.  With end-to-end being 
a possibility, I could 'route around' the cache - this is possible in 
the protocol.  But when end-to-end is not possible, as in the case of 
a NAT bottlenecking all traffic, then I can't route around the old 

Designing DNS to be tolerant of NATs is a pain, as Keith alludes to 
in a previous message.  It's remotely possible, but a real waste of 
time and energy.  Just as other applications want to mimic DNS's 
recursive lookup behavior (which is a mistake), DNS shouldn't turn 
around and mimic store and forward behavior of its messages.  The rat 
holes are already known, fortunately no one is trying to measure 
their depth. ;)
Edward Lewis                                          +1-703-227-9854
ARIN Research Engineer
Received on Friday, 6 December 2002 09:18:19 EST

This archive was generated by hypermail pre-2.1.9 : Tuesday, 24 February 2004 19:46:25 EST