Dave,
A lot of what you wrote (and I don't quote below) has indeed been
said many times. However, I think you left unsaid a few important
things that have also been said many times in the past.
So, the remaining problems [other than routing table growth and
traffic engineering] of having IP Addresses be used as identifiers
are not IP's. They belong to other areas. (Or, rather, Areas...)
What was repeated at least 7 times during this morning's INT are
meeting is really a sad piece of history:
The majority of today's applications use some derivative
of the Berkeley Socket API. The only thing the Berkeley
socket API understands (in practice) is IP addresses.
The TCP TCBs are also bound to IP addresses; something that,
based on our current success, must have been a very good
design decision back when it was made.
So, if you say that the other problems (like mobility or the use of
IP addresses within applications and management systems) belong to
other Areas, you are basically stating that both TCP, all
applications, and most of our current management systems must be
changed since they have been implemented wrongly from today's point
of view. Now, I might buy (but don't) that as an architectural
statement, but realistically, if our job is to keep the Internet
running even when the world is changing, I'm afraid that we do and
will need backwards compatibility. (See Section 3.3. of draft-
nikander-ram-ilse-00.txt)
[Footnote: The reason why I don't quite buy your argument even
architecturally is that I believe that for most functions there is a
natural level of granularity (e.g. subnet, host, or application)
where they apply. If we are able to implement them within the
protocols that act on that granularity, we can gain in terms of
simplicity, protocol efficiency, and security.]
For example, multi-homing is naturally a transport problem -- and
it is fine if the solution is embodied as a shim above IP, since it
also can be characterized as a shim *below* transport...
I would only agree if there was a clear distinction that the IP layer
works in terms of interfaces and it is the transport that works in
terms of hosts (and not even then for site multi-homing). But that
is not really the case. The legacy transport (TCP and UDP) have no
concept of a host. Architecturally (and I am referring back to
Salzer), the architecture has been missing, from the beginning, host-
granularity identifiers.
--Pekka Nikander
_______________________________________________
Int-area mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/int-area
