The most critical operational need on the internet is a global
routing system which will scale to orders of magnitude more
richly connected end sites.

It is believed that computational constraints imply that this
must be without all well-connected nodes in the graph having to
know global state.

Traffic engineering, VPNs, etc. are mandatory parts of the
environment, but routing may not need to be the [only] means
by which they are achieved.

The IETF and other venues have gone nowhere for over a decade.
Centralized schemes derived from seeing only the operational
philosophies of one warped telco's network are unrealistic.

Clueful folk such as Sean Doran suggest that the p2p overlay
networks are some of the only ones thinking out of the box.
But they seem to be at war with the physical topology as
opposed to enjoying cooperative mutual exploitation.

Tim Griffin's metarouting proposes language to express new and
well constrained routing protocols, but gives us no help with
the critical issue of designing protocols for scalability.  But
it might be great fun to describe and test new protocols using
metarouting, VINI, etc.

8+8/GSE and the locator/identifier separation work, while
giving us more opportunity for topologic information reduction
does not address the cause of over half the information in the
routing table today, intentional de-aggregation for traffic
engineering and a false sense of security.  Formal routing
security should reduce the latter.

The layer two and three networks are overlayed on layers zero
and one.  Operational experience has taught us that when these
overlays are not congruent with the layers on which they ride,
management and troubleshooting issues are greatly exacerbated.
Above layer three, overlays seem to be either
  o research platforms for those who do not wish to spend a lot
    of money on layer zero/one networks on which to run
    experiments
  o p2p and other deployments which are forced to overlay as
    they are at traffic/routing war with those who own layer
    three.
It is not clear how these forms of overlay routing networks are
actually useful or efficient as production networks once we
remove the p2p war issues.
  o Skype, SIP, and other application layer overlays are not
    really routing lessons, but applications whose paths are
    performance sensitive.  This is traffic engineering.
  o Layer three VPNs, such as IPsec, are not routing overlays.
    While layer two VPNs are neither private nor do they really
    care about routing.

We desperately need routing architects and researchers to break
us out of the closing walls of scaling and complexity.