[2617] in Commercialization & Privatization of the Internet
AUP
daemon@ATHENA.MIT.EDU (Peter S. Ford)
Mon Mar 23 07:06:31 1992
Date: Mon, 23 Mar 92 05:05:05 MST
From: peter@goshawk.LANL.GOV (Peter S. Ford)
To: com-priv@psi.com
I believe most people will agree that the microphysics of hop by hop
destination based packet forwarding makes enforcing AUPs which talk about
end to end flows hard to enforce thru routing. ( Someone should write this
up for the annals of the Internet. Anyone???)
It is important to note that this does not mean that this is an
unsolvable problem, and in fact, one can look at work like InterDomain
Policy Routing to provide a clue as to what we might expect in the
future. It is also possible to restate the AUPs, but I believe these
new AUPs would be even more restrictive, and most people would
probably reject them as a step in the wrong direction.
I believe we are seeing stress due to the application of
technology which was designed, tested and deployed to serve a single
community of interest, when in fact the current Internet is far from a
single community of interest. In the former case, if there were
multiple paths between two sites using any or all of the paths was
okay. In the later case, it may not be okay, but that in and of itself is
okay.
Can we agree that it is unlikely every network in the Internet
will want to be a transit for every other network? Let's run the
thought experiment:
NSF lifts the AUP. (Sorry Steve, but perhaps we can get you
a minimum security cell somewhere nice in Maryland
with an Internet tap so you can still do e-mail. :-))
Now where does traffic transit? I bet a large amount will be going
to the NSF backbone.
Now what is our favorite topic on com-priv? Can you say
settlements? Can you say "I don't want to be the
1990's version of ihnp4!"?
I don't see the current formulation of the CIX solving the
problem of transit today. It does a great job of solving the access
problem as long as consenting networks are willing to eat arbitrary
transit costs. Perhaps this will be a natural state when the cost of
carrying photons reaches its limit of zero, but we are not there
today. (Note, I am rejecting the notion of a single interconnection
point since we lose some nice properties, like short paths.) I can
imagine some alternatives they (or other CIX-wanna-bes) might be considering:
settlements
a shared transit system between CIX points, somehow jointly
funded by CIX members (I bet this ends up looking like
some kind of settlements scheme)
sender keeps all, what does this mean for the regionals and the
global network providers ...?
I am sure there are other schemes that I have not listed here.
It is even possible something unexpected might evolve, much like
the rampant hub and spoke evolution of the airlines since they
were deregulated. (I hope I do not end up as the Salt Lake City
of networking where a single provider can charge whatever it
wants due to a monopoly position!!!)
[note all of the above may require some notion of (eeeek!)
accounting.]
What is my point? Much of the NSFnet AUP argument is unproductive
because it simply takes the reader from one set of assumed evils
to another set of assumed evils. As it stands, NSF funds a network to
support research and education, I think this is a good thing. This network
interconnects with many other networks some which serve communities other
than the R&E community, again a good thing. They interconnect so
that academia and research sites can communicate with non-R&E
people for purpose of knowledge exchange, tech transfer, etc., a good
thing. NSF would like the bandwidth/service it is paying for to support
connectivity to and/or from R&E sites only, thus helping to insure the
gumint is not losing its shirt, a good thing.
Current tech is not so great at selecting between two
available paths based on some criteria (policy, bandwidth, latency,
reliability, etc.). Some people propose the solution to the problem
is to get rid of the AUP and subsequently not having NSF fund any
particular network. I suggest the solution is to apply better
technology, in other words, get a new physics. This may include some
rational notion of accounting. For the forseeable future, the gumint
will want to target special networks to special problems (NASA
missions, inteconnecting supercomputer centers, interconnecting K-6
schools, etc.), and there will be AUPs. I would also argue that in the
future, many commercial ventures will want to have policies which look
a lot like AUPs. Time for new tech, new forwarding micro physics,
more bits ... On the other hand, it is probably also time to have
rational discussion of settlements and the like ...
cheers,
peter
P.S. Yes, I am sending this to kick the discussion out of the bash
the NSF AUP chaotic attractor. What about that Pennsylvania
Caller ID decision? What does it mean for src/dest routing and
recording in packet headers ...
