[rrg] Constraints due to the need for widespread voluntary adoption

[Robin Whittle <[e-mail]>]

Lixia wrote:

  http://www.ietf.org/mail-archive/web/rrg/current/msg05231.html

that from 15 December to 15 February we will be discussing proposals,
and after that writing the recommendation.

In the two weeks to then, I would like to prompt a discussion of the
constraints which apply to any solution to the routing scaling problem.

I wrote about this in April and updated the text after some helpful
and generally positive feedback:

  http://www.firstpr.com.au/ip/ivip/RRG-2009/constraints/


Does anyone agree with some or all of this?

Can anyone suggest improvements to the text?

Can anyone imagine a way the routing scaling problem - today's IPv4
problem or the same problem for IPv6 in the future - could be solved
without being subject to constraints such as these?


As far as I know, the solution needs to be voluntarily adopted by the
great majority of networks which want or need multihoming,
portability of their address space and/or inbound traffic
engineering.  We can't impose the solution.  We need most - ideally
all - networks which already, or in the future, want these things to
voluntarily adopt the solution.  Otherwise too great a proportion of
the growing number of networks will continue to use the existing
unscalable BGP techniques.

Scalable routing beyond simply dealing with the needs of the sorts of
networks which actually choose to use PI address space today: that is
to provide one or more scalable techniques which a much larger number
of networks, typically smaller ones, can adopt to achieve these
goals.  So the voluntary adoption constraints apply to both the sorts
of networks which today have PI space, and also to a much larger
number of typically smaller networks.

The most obvious set of new adoptors of multihoming etc. would be
businesses and organisations who wanted robust Internet connectivity
to their offices, factories etc. via two or more physical links with
different ISPs.   The first link might be DSL or fibre and the other
might just be WiMax or 3G or some other wireless technique - so it
doesn't necessarily involve two cables.  The aim may not be to
maintain normal bandwidth, but to maintain enough connectivity for
mailservers, SSL web servers, VPNs, video-audio conferencing etc. to
continue business activities in the event the main link dies.

I estimate an upper limit on the number of such networks by assuming
that for every 1000 people, there is only likely to be one such
organisation, typically a business, with sufficient motivation to get
two links.  If we assume a maximum population of 10 billion people,
then that means there would be no more than 10,000,000 such networks.
 I assume that for most homes and businesses the reliability of their
single DSL or wireless service is acceptable.

As far as I know, the only scenario in which more than about 10M
networks must be accommodated in the scalable routing system is where
each mobile device has its own piece of address space, such as in:

  http://www.firstpr.com.au/ip/ivip/TTR-Mobility.pdf

That could, in principle, mean billions of separate pieces of address
space, such as 10 billion people each with a "cellphone".

I suggest we leave mobility out of the discussion about voluntary
adoption, since we don't need worry about mobility when solving the
foreseeable routing scaling problem, and because if such mobility
techniques are developed, there will be no problem with voluntary
adoption since those techniques will be the only way to do it.  There
 is no danger of cellphone users using existing BGP techniques to
advertise their prefix in the DFZ.


  - Robin