[Discuss] Disruptive Deployment of New Network Design

[J. C. Jones]

Hello,

Many have said that the biggest obstacle to IPv6
deployment has been the intransigence of ISP's who saw
little to be gained from such deployment.  Others have
said that a goal of GENI is to allow multiple
alternative protocols to compete on a world-wide basis
in a research environment. Perhaps, in addition to the
research-oriented model that GENI proposes, it would
be useful to discuss the possibility of a new, more
immediate (< 5 years) commercial model where a new
network protocol stack is injected into the live,
commercial Internet.

We can see from P2P networks like Bit Torrent and
Kazaa that it is possible to install communication
engines on millions of computers around the world. 
These communications engines, for all practical
purposes, are linked to each other via network links. 
If QoS is ignored, these virtual links could be
regarded as real physical links with reduced capacity.
 So it is conceivable that an entirely new Internet
protocol stack and network could be created from such
machines. One could imagine that IPv6 could have been
deployed in this manner - we could have used current
IPv4 UDP packets as the link-layer frames for IPv6. 
Provided that there was sufficient virtue in the new
network model that warranted new application
development on that model, I suspect that users would
have been willing to overlook the performance
reduction, at least long enough for the new network to
take root, after which the performance could be
incrementally regained by dropping the links from
virtual to physical.

I would like to know if anyone thinks this is
feasible, as well as answers to these questions:

1. What is the true necessity of buy-in from the ISP's
before global deployment could be achieved?  
2. Is it true that users will ignore a performance
penalty of any overlay if their major headaches could
be solved?  If UDP were used as link-layer frames,
what would you say would be the reduction in
performance? 10%? 20% 30%?
3. Will users try a new distributed application
without regard to the underlying network as long as
the application fulfills the promises it makes?
4.  Do ISP's hold a monopoly on all "core" machines,
or are there non-ISP entities that own machines in the
core that could support a new network protocol model
in the context of, say, multicasting?
5. Is it necessary for IPv4 to die before a new
protocol model could gain traction?
6. Assuming the security and mobility problems were
solved, how long would it take before the new network
model became recognized as a viable separate network? 
What would users call this new network?
7. If a new network were to take hold, assuming that
it had a feature set that was essentially a superset
of that provided by Internet-IPv4, with what
difficulty would existing applications (web browsers,
etc.) be recoded to take advantage of the new network?
 Would it be possible for such applications to support
both network models simultaneously during a period of
transition?
8.  What happens if the new network contains a gross,
intolerable error, say, a bad choice of cryptographic
primitive used to enforce security?  What would be the
feasibility of a large-scale deployment of updates?
9.  Is there any commercial value in being a prime
provider of the new protocol stack?  Would network
router vendors feel compelled to implement the stack
in parallel with IPv4/6 to accommodate the demands of
their customers?  Will ISP's feel compelled to
purchase such hardware from hardware vendors if there
is potential for revenue generation from, say,
multicasting?

10. Is there anyone else who feels that this method of
introduction has merit?

-JC-


 
____________________________________________________________________________________
Do you Yahoo!?
Everyone is raving about the all-new Yahoo! Mail beta.
http://new.mail.yahoo.com