North American Network Operators Group
Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: Fundamental changes to Internet architecture

  • From: David Andersen
  • Date: Fri Jul 01 11:28:28 2005 
On Jul 1, 2005, at 9:40 AM, Eric Gauthier wrote:

Dave Clark is proposing that the NSF should fund a new demonstration
network that implements a fundamentally new architecture at many levels.
Not that I want to throw any more fire on this, but I think the article is
talking about National Lambda Rail. From what I've seen, this is supposed to
be a next-generation Internet2 based on wavelengths instead of pipes. I don't
know a lot about it but, from what I've seen, my impression is this. Keep in
mind that I'm not really involved with the NLR stuff directly, so my thoughts
are really as an outsider looking in.
To clarify a bit: Dave Clark is talking about a new, proposed research agenda for networking research that emphasizes the heck out of making the research become real and relevant. He's not talking about building an NLR or an Internet2, though both NLR and I2 are resources that can and probably will be used as a part of the demonstration network, if the project really takes off.

In fact, Fergie's later comment "... We're pretty far along in our current architecture to 'fundamentally' change" is actually the root of what I think DC is trying to get at. I think it's a very reasonable question to ask: Is the Internet heading towards a local maxima? (I don't know the answer!) What is it possible to change in today's Internet? Imagine a couple of things that seem desirable:

If research came up with an improved inter-domain routing protocol that had faster convergence, better security and better stability than BGP, but that was unfortunately in no way backwards compatible, could we deploy it?

A solution to DDoS that required another change to the basic IP packet format?

An improved intra-domain management and control system?

Of those, some seem possible -- particularly the latter, given that it could be deployed by a single ISP on its own, giving it (ideally!) a competitive advantage over others. A BGP replacement, if the designers/ietf/etc. couldn't figure out a way to make it backwards compatable? Not so sure. Another IP packet format change, after all of the pain of trying to get IPv6 deployed?

Perhaps more of the answers to these questions would be "yes" if it were possible to demonstrate - at scale - that the new protocols were actually effective and worthwhile. Or perhaps the answers would be "yes" if that demonstration network exploded in popularity because it had those features, and the NSF found itself with another Internet on its hands. :)

I think it's these kind of questions that Dave Clark is trying to get at, much more than just trying to build a really fast demonstration network.

Is the clean-slate approach the way to go? I don't know. It could work out well, or perhaps academia would be better served by sending more of our students to summer internships at ISPs who're doing innovative things. I do know that Dave Clark is a damn smart guy, and he does have TCP under his belt loops. Sometimes you have to aim for the sky...

Disclaimer: While I've heard some of the discussion about this proposal, I'm speaking only for myself on this one.

-Dave