more on Growth of the Internet May Take Nothing Short of a Revolution

[Dave Farber <dave () farber net>]

Delivered-To: dfarber+ () ux13 sp cs cmu edu
Date: Tue, 23 Dec 2003 22:35:50 -0500
From: Bob Frankston <Bob2-19 () bobf frankston com>
Subject: RE: [IP] Growth of the Internet May Take Nothing Short of a 
Revolution
To: dave () farber net, ip () v2 listbox com
Cc: "David P. Reed Ph. D." <dpreed () reed com>,
 'Dewayne Hendricks' <dewayne () dandin com>

[Dave -- I'm sending this note with some hesitancy because I feel I should
provide more detail and have a specific proposal but given the WSJ story it
seems appropriate to respond with an overview of what it really means to
rethink the Internet -- it's far more about being dynamic than speed or
other technical demos.]

Revolt? More like just create the kind of Internet we need out of the pieces
lying around. After all, it is just a prototype.

I've been thinking a lot about these issues and have decided that my
"dotDNS" (removing the semantics from the DNS) and IPV6 are not enough. A
few years ago David Reed and I talked about rethinking the Internet now that
we know how well it works. I put it to the side because I thought it would
better to take incremental approaches first.

But the incremental approaches failed to address the fundamental issues. In
speaking to Paul Vixie and others it became apparent that even if I got what
I wanted the routing problem was going to be a limiting factor. In speaking
to Dewayne, it also became apparent that radios are a forcing issue for the
not-very-dynamic Internet - once we aren't dealing with spectrum management
then radios can simply come into existence and disappear in less time that
it takes to petition for an IP address.

I'm working on a detailed proposal but the basic idea is very simple --
going back to basic end-to-end model. Start by having the concept of end
point but now that the idea has proven itself we don't need to associate it
with a physical computer. We can keep the concept abstract. We can also get
past tying the end points to any particular model of routing. In fact we can
have abstract relationships that don't require routing. This concept is
already evident in the use of URNs for their uniqueness without associating
them with an actual device.

The End Point Identifier (EPI) can be a very long random number that is
self-chosen, un-guessable and usable as crypto key.

Routing is just another service and not a layer. Unlike the classic IP
address the EPI is fundamentally stable. The routing infrastructure doesn't
have to track them -- it just needs to deliver packets to a nearby way
station which is what it already does but only a special case -- the
gateway. An end point can be a conversation that uses a particular computer
as a way station for routing. The conversation can move to another machine
without disruption because it is defined by the EPIs and not the computer of
the moment. The host computer isn't an issue, NATs are not an issue and a
particular provider's routing addresses are not an issue.

Since the EPIs are all first class addresses and decentralized there is no
need for the DNS to provide stable handles nor ICANN et al to dispense
identifiers.

Crypto is necessary in order to avoid dependency upon the benevolence of
every way station along the route. WEP and all that are also nonissue.

Obviously I'm only touching upon the concepts here and will be writing about
this in more detail as well as working with others to drill down on the
design details. Even if people don't accept my contention that it is very
doable the very idea of getting past all of those issues should be
interesting if not exciting.

I realize that this short explanation will probably confuse a lot of people
but that's the risk of giving a short synopsis of ideas whose simplicity is
based on underlying principles that take time to absorb. In reading the WSJ
article I'm amused by the idea that we should resurrect the concept of
circuits -- people are confused anyway.

The real power of this approach comes from becoming fundamentally and
intrinsically decentralized. It can be implemented by those who choose to
without asking permission. The current Internet provides an initial set of
routing options so there is no need to build a new physical Internet just
for the new protocols.

Sometimes it's easier to repave than keep patching. It is far simpler to
just assume stable names and use whatever routing is available than
requiring even more complex routing. It is infinitely simpler than dealing
with the ICANN debacle or hoping the DNS works or hoping your operating
system does V6 at all let alone with encryption.

-------------------------------------
You are subscribed as interesting-people () lists elistx com
To manage your subscription, go to
 http://v2.listbox.com/member/?listname=ip

Archives at: http://www.interesting-people.org/archives/interesting-people/