[e2e] where end-to-end ends

[Eric A. Hall]

Eric Travis wrote:

> I've always interpreted the ultimate "end" in
> end-to-end to be the communicating processes
> rather than the transport entities who are
> providing services.

The end-to-end principle is marked in comparison to the gateways which
preceded it, and thus hinges on the concept of a tuple between two
application end-points. Whereas older and/or private network architectures
required the use of "smart" intermediaries, the Internet model allows
"smart nodes, dumb network", and thus the end-points communicate directly.
This makes the tuple the watermark.

> Now I'm curious - am I really bound to having only
> a *single* layer-4 hop before "end-to-end ends" in
> the absolute sense?

> All this is really no different than e-mail. Wouldn't
> you agree that e-mail correspondence between two
> people is still legitimately labeled as "end-to-end"
> communications?

There are many protocols which insert multiple e2e hops into the middle of
a user session, but those individual hops are there for the benefit of the
application protocol and could be removed. EG, I could configure my
workstation to send mail directly to your mail server instead of letting
SMTP store-and-forward the mail through multiple intermediaries. Or I
could remove my local caching recursive DNS server and let my workstation
try to find you directly. And so forth. Those intermediaries are there for
the benefit of the apps.

Moreover, the individual hops still behave as end-to-end, because they
still use tuples. This is despite them being store-and-forward, in-line
caching, proxied, or whatever. End-to-End ends at the last reachable
Internet-compliant system, regardless of protocol design efficiencies.

The more interesting question I think is where does "the Internet" end.

NATs mark the end-to-end boundary, for example, since they are the maximum
possible termination point for Internet-compliant systems. No system
behind the NAT is reachable via end-to-end then it is not compatible with
Internet protocols and operational principles.

But does the end-to-end edge also mean that all of the hosts behind the
NAT are not "on the Internet"? or does "the Internet" refer to a broader
collection of networks which are not necessarily governed by end-to-end
principles? By that broad scope definition however, are protocol gateways
that are not IP-IP also considered parts of the Internet? Are cell phones
which are only reachable via gateways "on the Internet"? Are they "on the
Internet" only when they are exchanging data with Internet hosts (assume a
proxy is involved), and "off the Internet" when they use non end-to-end
principles internally?

We know where end-to-end ends. Where does the Internet end?

-- 
Eric A. Hall                                        http://www.ehsco.com/
Internet Core Protocols          http://www.oreilly.com/catalog/coreprot/