[arin-ppml] Map-encap for better utilization of IPv4 space?

[Robin Whittle]

Hi Per,

In "Re: [arin-ppml] Portable address space vs. IPv6 auto-numbering",
you wrote, in part:

> If continued growth of the net through re-allocation of unused
> resources is the no-brainer you claim it is, why haven't the RIR
> yet performed and published results from surveys to document it?

I know very little about RIRs and address allocation policy - and I
never wrote it was easy.

It is difficult to do now, without generating far more BGP
advertised prefixes, generally longer prefixes, as space is divided
more carefully in smaller chunks.  That drives the routing scaling
problem, so it is strongly resisted.

The pressure to use IPv4 space more efficiently will grow and grow,
since I can't see how IPv6-only services could suit most end-users.


> How much of the previously allocated space can be expected to be
> available for re-use given various conditions (prices and/or
> political pressure), from either registered LIRs or
> legacy-holders. Is it possible to meet the market's demand for
> sustained growth for any significant time past depletion of the
> free pool?

My understanding is that only a relatively small fraction of the IP
addresses currently advertised are actually in use:

  http://www.isi.edu/~johnh/PAPERS/Heidemann08a.pdf
  http://www.firstpr.com.au/ip/host-density-per-prefix/

perhaps 200 to 300 million.  I know ping surveys have their
problems, but the figure I and the USC/ISI team got was around 110
million.

A map-encap system - LISP, APT, Ivip or TRRP:

http://www3.tools.ietf.org/group/irtf/trac/wiki/RoutingResearchGroup

would enable a lot of the space to be finely and arbitrarily divided
amongst end-users without adding to the routing scaling problem.  I
figure this could make much better use of IPv4 space in general.
Each such end-user network could get portable, multihomable space
without the costs of current PI space, and without having to grab
more than they need, since the map-encap systems can easily slice to
1, 2, 4 etc. IP address prefixes.

It is a big question which I can't answer in detail, but if the
world is getting by now with a few hundred million IPv4 addresses in
use, out of 1.7B advertised, and there are 3.7B which could be
advertised, then I reckon that a map-encap system gives a great deal
of scope for much higher rates of utilisation and therefore many
more years of life for IPv4 without any end-user networks really
being unable to obtain IPv4 space (perhaps in smaller than ideal
chunks) for a reasonable price.

I am not sure this directly helps ISPs who don't want map-encap
space, but want swaths of space to connect DSL customers, each with
an IPv4 address.  It is more that the map-encap scheme and the
pressures on end-user networks to get their own, multihomable,
public, portable PI space will drive the use of map-encap and so fit
far more end-user networks into a given amount of space than is
possible now.

  - Robin              http://www.firstpr.com.au/ip/ivip/