On 9/26/02, Lloyd Wood wrote:
On Thu, 26 Sep 2002, Caitlin Bestler wrote:
So, as originally proposed an IP fragment is a fully
self-routed L3 datagram.
well, not self-routed; you need routing state. I don't
think the difference between routing table state and
circuit-switched state is all that great; anything beyond
hot-potato is fundamentally stateful.
Given the source interface, the *meaning* of an IP header
is not supposed to be dependent on the routing tables.
The routing tables merely implement that meaning.
By contrast, the meaning of an ATM circuit is dependent on
the context in which it was established. There is no
expectation that there is any meaning to this circuit
identifier beyond those imparted when the circuit was
I would maintain that all the IP exceptions that leap to
mind, such as NATs and load-balancing switches, do not
actually violate this. The intended destination is merely a
In any event, regardless of similarities in execution, the
difference in intended semantics is valid and fundamental.