[e2e] a means to an end
David P. Reed
dpreed at reed.com
Thu Nov 6 13:29:29 PST 2008
The original "six degrees of separation" experiment showed that networks
can route messages without being told "where". The only argument is one
of efficiency, and that argument only applies when a) nothing moves very
fast, and b) the sender can be assumed to know "where".
Let's say I have a message for all Republicans. I can send that
message quite reliably, without ever knowing "where" the set of all
Republicans is. In fact, it's a feature that I can do that. What's
particularly nice is that in some cases, it is far *more* efficient to
do that than to "first, find the locations of all Republicans" and
"second, deliver the message to that set of locations". Maintaining a
complex location set as an intermediate computed value seems like it
would cost a lot more than just telling each Republican I know to pass
the message to any Republicans he/she knows.
OK, OK. that's not communications - because only the holy scripture of
the Cisco router defines "real" communications. :-) They have a patent
on it, which they license to their IETF.
Fred Baker wrote:
> It's kind of hard to get a message to a destination, fixed or mobile,
> without giving the network an idea of where it is to go. That's kind
> of the point of the locator.
> That said, the point of the endpoint ID is to identify the application
> on the system, the transport connection endpoint, independent of its
> location. If the application moves from one system to another, the
> transport connection needs to be able to follow it.
> Gee whiz. RFCs 1483, 1753, 1922, 2102, and 2103.
> On Nov 6, 2008, at 12:06 PM, David P. Reed wrote:
>> Why should "location" be relevant to networking? Must all wires be
>> buried permanently in the ground? Does wireless and mobility not occur?
>> Michael Welzl wrote:
>>> On Wed, 2008-11-05 at 13:26 +0100, Ali Ghodsi wrote:
>>>> Jon Crowcroft wrote:
>>>>> but, BUT this is sidestepping the big problem
>>>>> which is to have a service which hosts the id<->loc mapping, which
>>>>> actually scales to the size of the expected workload...
>>>> Can DHTs be part of the solution, and if not, what are the
>>>> essential features which they are lacking? (trying to fish for
>>>> research problems)
>>> Location dependance, which even the "id" part should
>>> have (at least in the form of some concept of "nearness").
>>> This was pointed out quite convincingly (IMO) by John Day
>>> in his "patterns in network architecture" book.
More information about the end2end-interest