[e2e] end-to-end -ness defeats content controls

[David P. Reed]

The canonical reference to the end-to-end argument is the original paper 
which I co-authored with Saltzer and Clark.  It existed in several forms; 
here are the most accessible:

Saltzer, J.H., Reed, D.P., and Clark, D.D., "End-to-end arguments in 
computer systems design," ACM Transactions in Computer Systems 2, 4, 
November, 1984, pages 277-288.

Online version: http://www.reed.com/Papers/EndtoEnd.html (for those who 
believe that print versions are becoming less "archival" than digital ones).

There are more recent references to the "policy" impacts of end-to-end design.

For example, there are a couple of related papers by Dave Clark and Marjory 
Blumenthal (write to ddc at lcs.mit.edu for references) on the places where 
the end-to-end approach conflicts directly with more centralized techniques 
and definitions of policy controls - and in particular they comment on 
areas related to the IPR issues you discuss.

[begin side comments, relative to my personal views on end-to-end design 
and IPR]

My personal opinion is that the term "intellectual property" itself 
distorts the argument in a way that is hard to avoid, and potentially 
dangerous.  It is only very recently in history that copyrights, patents, 
and trade secrets were labelled (by some) to be "property".  The legal 
history of these things started out without the concept of property.  A 
patent is a right to block certain *uses* of an idea (not ownership), a 
copyright is a right to block the *copying* of an expression (not 
ownership), and a trade secret is a right to enforce the binding of people 
to maintain secrecy.  None of these rights make bits into property.  The 
right to block may be assigned or sold (but the underlying idea is not thus 
assigned or sold), so the "right" behaves like property, but the bits 
don't.  All of these things are *limited* in scope by time and other 
constraints.  Again not *property*.

So a proper end-to-end design process would focus on means to achieve 
effective enforcement of the *rights* above.  By creating "property-like" 
packages of bits, designers are overreaching, and participating in "making 
new law".

One of the benefits of end-to-end design approaches is that it forces 
designers to think through what the proper goal of a designed function 
should be.  Too often, designs are driven by the tools/techniques available 
"in the network".  Those tools get subtly converted to become the goals by 
pragmatism.  If the tools "overreach" (implementing a far-too-rigid and 
extensive form of the desired function) there will usually be some who 
support that overreaching as a goal.

For example, it is by no means clear that "rights management" of the form 
described by law and precedent can be implemented in the network.  But it 
is certainly true that much more onerous controls can be implemented in 
some areas of the network (viz. Microsoft's latest approach to 
anti-circumvention in its Windows Media architecture).  Do those controls 
merely "optimize" enforcement of existing laws about copying?  Or do they 
create vast new areas of capability to interfere with actions other than 
copying?  From the marketing literature for these new digital rights 
management schemes it is clear that they have created many new controls: 
for example specifying which application programs are allowed to process a 
particular expression, and extending the "term" of copy blocking to a term 
longer than the limit mandated in the law.

Sadly, both lawyers and system designers participate in the overreaching, 
and in the word-spinning that they use for political cover.  Equally sadly, 
some of the folks interested in circumvention use incorrect technological 
arguments to claim that the old laws should not apply even  in cases where 
it is clear, rather than using the political process to get the laws changed.

[end of personal comments]



At 11:14 AM 10/1/2001 +1000, George Michaelson wrote:

>I just wrote a very wanky paper which is certain to be rejected but the
>central proposition was that end-to-end -ness and some related concepts
>like encapsulation of protocols make attempts at content control completely
>specious.
>
>Would fellow e2e-ers feel I overstepped the mark? It seemed like a small
>line to me:
>
>         1) encapsulation as a generalized concept implies that if you
>            can achieve end-to-end exchange of data, you can overlay an
>            unconstrained IP tunnel at costs in speed and bandwidth.
>
>         2) port bindings for service are meaningless tools to constrain
>            dataflow, the trivial proof is to edit /etc/services and 
> inetd.conf
>            and run SMTP on shifted ports. Or, to run some other service 
> on 25.
>
>                 [yes, it defeats global governance, but in context for
>                 a smaller n-way community of interest it works. I could
>                 have said BXXP or other multiplexing could also apply so
>                  port 25 is SMTP if helo works, but falls back to other 
> stuff]
>
>         3) convergeance is commoditizing the edge-device to include IP anyway
>
>         4) because any device can be edge, core, route, server or client the
>            idea a hierarchy of control exists is broken.
>
>Am I wide of mark here? I felt that e2e was a good over-arching design goal
>which was so fundamental in the Internet that it acted as a general principle
>to bind these discrete ideas into a commonality of 'you can't constrain it'
>type rules.
>
>Do we have a canonical reference to end-to-end I can cite?
>
>cheers
>         -George
>--
>George Michaelson       |  APNIC
>Email: ggm at apnic.net    |  PO Box 2131 Milton QLD 4064
>Phone: +61 7 3367 0490  |  Australia
>   Fax: +61 7 3367 0482  |  http://www.apnic.net