[e2e] e2e principle..where??....

[Eric A. Hall]

"David P. Reed" wrote:

> The "transparent cache" approach (pioneered by <?>) where the
> network itself spoofs HTTP and tries to second-guess what the user
> wants - that is not end-to-end, and creates huge secondary problems

And then there are NATs, which are e2e for the external link but which are
transparent end-points for the client link. ugh.

While it's apparent where the "ends" for any given session lie, these
scenarios do raise some interesting questions about the e2e model.

For example, is clean layering a fundamental requirement of the e2e model?
Whereas the historical usage has been that "the app which is bound to the
socket (and which is processing the layer-7 data) is the end system", the
socket does not represent the end system in these scenarios. Does it
matter? Should it matter?

Or on a broader question, has the nature of the Internet changed so much
that e2e is an artifact of an older network model? We have also seen that
the network is no longer "distributed", but that it is nowadays mostly
decentralized and hierarchical due to the economic realities of building
and managing high-capacity networks. Is e2e a similar artifact?

Note that e2e can still exist, but maybe it doesn't exist by default
anymore. For example, Internet principles such as distributed and e2e
allow end-systems to route around roadblocks in the network, and users can
setup their own links between end-points if they need to do so (just put a
modem bank on every web server), so the distributed topology and e2e
concepts still exist in valid concept form. But do they exist as default
on the Internet anymore. Probably not.

Should they? Is this a necessary thing for the Internet to continue?
Probably not. There are efficiencies from breaking e2e (caches, proxies,
store-and-forward, all have shown this), just as there are efficiencies
from ignoring the distributed model. Does this mark the end of the
Internet, and does this make the network something else. Does the Internet
even exist anymore?

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