On Aug 20, 2007, at 3:16 PM, Hallam-Baker, Phillip wrote:
Which is what prompted the original point I made in the plenary:
when someone is using the end to end principle to slap down some
engineering proposal they don't like I would at least like them to
appear to have read the paper they are quoting as holy scripture.
hear, hear...
The Internet has changed and we need to recognize that there may be
different circumstances that lead to the desirability of a
different approach. My belief is that the most appropriate
architecture for today's needs is a synthesis which recognizes both
the need for the network core to be application neutral and the
need for individual member networks of the Internet to exercise
control over their own networks, both to protect the assets they
place on their network and to protect the Internet from abuse
originating from or relayed through their network.
The funny thing is that I'm not convinced that this is a change.
As I have said in this and other fora, there is an argument for
functional complexity in the network, and routing is its poster
child. Routing could be done from the edge and the network be unaware
of how to route across it. It would be very difficult for a service
provider to make any guarantees to his customers in such a case, but
the obvious algorithms one might use are well known. We don't do that
- we do indeed treat routing as an acceptable form of complexity that
leaves significant state in the network - specifically because we
find commensurate value in the functionality provided.
RFC 61 (August 1970), quoting a paper by L. Roberts and B. Wessler in
the same year, observes that
"A resource sharing computer network is defined to be a set of
autonomous, independent computer systems, interconnected so as to
permit each computer system to utilize all of the resources of each
other computer system. That is, a program running in one computer
system should be able to call on the resources of the other computer
systems much as it would normally call a subroutine."
Here, the author is clearly thinking of processes on computers. But
just the following year RFC 164 reports that the Air Force wanted to
build an autonomous network, in 1979 the SRI NIC commented (RFC 756)
on the need for autonomous name servers in various networks (a line
of reasoning that led to the present DNS a few years later), and in
1983 RFC 820 reported the assignment of the first Autonomous System
Number, used by RFC 827's EGP protocol for separating autonomous
routing systems at the edge from the BBN-operated ARPANET.
From my perspective, it has been at least 25 years, if not 37, since
we recognized "both the need for the network core to be application
neutral and the need for individual member networks of the Internet
to exercise control over their own networks". What we have struggled
with since is the recognition that while the core needs to be
application neutral, it isn't necessarily service-neutral; the
development of two QoS architectures, the traffic engineering in
MPLS, and the amount of bickering that has gone around declaring such
things to be inappropriate - ("just get enough bandwidth and all that
goes away") - has been all about the assertion that the core should
be application-neutral vs the assertion that it needs to be able to
offer specific services.
I read in the paper this morning that the "Web" is running out of
bandwidth. The article quoted Metcalf and others and their
predictions of the death of the Internet, and noted that Bob finally
blenderized his columns in front of an audience. My company figured
in the article. So did the advent of video content, which requires
significantly more bandwidth and significantly lower loss rates than
more traditional voice services or TCP-based applications. Had the
author had a clue, s/he would have said something about the Internet
once again changing its fundamental service, from enabling terminal
service, to moving files using FTP and Network News, to moving
smaller files using SMTP and HTTP, to basic audio services, to peer-
to-peer file sharing, and now to video. Not that the old goes away
right away, but we add on and the old becomes less important.
What we need to do is figure out how to let the intelligent network
core work cooperatively with the intelligent edge to let it do
intelligent things. Right now, the core and the edge are ships in the
night, passing and occasionally bumping into each other. No, we don't
want unnecessary intelligence in the core. But, as with routing, I
will argue that there is some that is constructive.
_______________________________________________
Ietf@xxxxxxxx
https://www1.ietf.org/mailman/listinfo/ietf