On 11/13/08 10:06 AM, Hallam-Baker, Phillip allegedly wrote:
I beleive that the question would not arise If we had a coherent
Internet architecture
The idea that an application can or should care that the IP address
of a
packet is constant from source to destination is plain bonkers. It
was
no an assumption in the original Internet architecture and should
not be
an assumption that any application should rely on.
That's not the problem. The issue is location. Once we have
established a session then how the packets are labeled for network
layer
purposes doesn't matter much (modulo security) but how do we get
communications set up in the first place? Suppose I want to reach
"foo". Who do I ask to find a locator for him? Split DNS works fine
when there are just two states, inside and outside -- a DNS server can
be configured to know how to respond in each case. But if you were to
sprinkle NATs all over the Internet there would be no place that could
give a confident answer about how I, over here, should name foo in the
network layer in order to get a packet to him, and have that answer
get
to me in the correct form. So it is very important to understand
where
we think it might be safe to put what kinds of NATs.
swb
