[e2e] New approach to diffserv...

[David P. Reed]

At 09:31 AM 6/17/2002 -0400, Jonathan M. Smith wrote:

>This whole discussion has been highly amusing.
>
>Let's just clarify some things that need to be clarified, perhaps with
>some help from the high priests of e2e on the list :-)
>
>1. Why is routing done with middleboxes?

It isn't.  A middlebox interprets data outside the IP layer.   Often based 
on presumptions and heuristics that may or may not be valid.


>2. Where is the financial incentive to build networks if the basic
>network architecture FORCES you to deliver a commodity with no value-added?

There's lots of value added in delivering data more reliably, with lower 
latency, jitter, and ability to reach more destinations more quickly.

Small rant: all businesses involve competition, except for those where the 
government grants monopolies (patents, rights-of-way, franchise 
laws).  People invest in building them all the time, on the theory they can 
deliver the service more efficiently than others.   The idea that no one 
would invest in a commodity business is so much horses**t, and is belied by 
the entire history business.    It is the case, however, that this line 
about "no one would invest in a commodity business" is often used in 
regulated industries (telecom, historically) to con governments into 
granting exclusive monopolies, capturing their regulators.   Whole 
generations of engineers then start to buy into this line.
What is true is that no one will invest in someone who provides a service 
that is more expensive than its competition unless they have some way to 
rape their customers (er, add value) by what economists call "rent 
seeking".   But networks have no special claim on "added value" - in fact, 
users would probably move off those networks that try to impose "added 
value" if there were no laws on every state and country's books granting 
exclusive licenses to rights of way, etc.


>3. Why is the network engineered in isolation from applications?

It's called modularity.   The network is useful for a wide variety of 
applications, most of which are not known, or their future value is not 
known.  Thus we do a general purpose network that allows a wide variety of 
uses.  Especially important because large parts of the economic value of 
networks comes in the form of increasing returns to scale and 
interoperation, so dividing the network up and optimizing each piece for a 
different use destroys value.

It's not fair to say it is engineered "in isolation", by the way.   In 
fact, the last network that was engineered "in isolation" was X.25, which 
ignored applications entirely to create a perfect virtual circuit 
architecture layered on packets.   The Internet had strong input about the 
general characteristics of computer-computer communication that differed 
from telephony's isochronous circuits.   But this was balanced by a strong 
desire to bind *any* particular application's requirements into the network 
- if we had observed the dominant use of the ARPANET (remote login to 
timesharing systems) and optimized around that, the Internet would have 
failed completely.


>4. Isn't e2e just a clever logical deception? It's of course obvious that
>an engineered artifact will have the maximum longevity it if it avoids
>any concession to current needs, but very few of us buy wheels, chassis
>and motor, instead opting for value-added services such as seats and a
>roof. E2e is just rhetorical "argumentum ad absurdum" wrapped up like
>some engineering mystique, no?

Read "Design Rules" by Baldwin and Clark.  It explains the economic linkage 
between modularity and economic success.
Following their approach one can see a lot of the value of the e2e argument 
(and Mark Gaynor did just this in his recent Ph.D. thesis). The end-to-end 
approach is an approach to maximizing the value of an architecture in the 
face of large market uncertainty.   It does not apply when the market 
requirements for all time are precisely known.   But in the case of 
internetworking, that does not seem to be anytime soon.  Perhaps when the 
world information economy is mature and unchanging in scale or structure - 
but is that 50 years or 1000 years hence?