[LINK] Crisis in routing and addressing

[Robin Whittle]

I am involved in trying to solve what is reasonably called a
"crisis" in the routing and addressing structure of the Internet.

More on the crisis here:

  http://tools.ietf.org/html/draft-iab-raws-report
  http://www.iab.org/about/workshops/routingandaddressing/

This is being discussed on the RAM and RRG mailing lists, as
linked to here:

  http://www.firstpr.com.au/ip/sram-ip-forwarding/#Links

I first came up with a RAM-based lookup scheme to greatly
improve the FIB (Forwarding Information Base) function of
Internet "core" (actually Default Free Zone - DFZ) routers - the
part which decides which interface to forward the packet to.
This turned out not to be the most difficult problem.

The real problem is that the routers and their BGP routing
protocol are gagging at the task of comparing notes about how
best to route over 200,000 different "BGP advertised prefixes" -
each one a division of the address space for which some border
router says "sent me all packets addressed to this range of
addresses".

This 200,000 number is growing alarmingly.  This is a problem
partly for the FIB functions of routers but also for the control
CPU and its memory.  Each CPU has to retain the "best route"
messages of all its peer routers, for every one of these
200,000+ subnets.  Despite these BGP messages being about as
terse as possible, the problem is getting out of hand.  Even if
routers were souped up with 64 bit CPUs with gigabytes of RAM,
there are concerns about the stability of the BGP system.

Unless some new architecture is developed, then natural growth
of the Net will require all the DFZ routers to be upgraded to
something overly expensive, complex and power-hungry, probably
every 3 years or so.  Since all Internet users pay collectively
for those routers, this is a problem for everyone.

I now have a new proposal Ivip (Internet Vastly Improved
Plumbing), which is an adaptation of a proposal LISP (Locator/ID
Separation Protocol) which has been discussed on the RAM and RRG
lists for a few months.

Like LISP, Ivip is intended to allow the Internet routing and
addressing system to scale to much larger numbers of end-user IP
addresses and subnets - which are portable (can be connected via
any ISP) and which can be used for multihoming and for traffic
engineering, without further burdening the BGP network with more
advertised prefixes.

Ivip should be incrementally deployable, whereas LISP requires a
new "Ingress Tunnel Router" in every edge network, unless hosts
in those edge networks are to be unable to communicate with
hosts using the LISP-mapped addresses.

My page for this is:

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

It links to the messages on the RAM list, which is currently the
only description of "Ivip".  I thought it up on Thursday night
and blurted it out to the list, rather than work in isolation on
it, and perhaps waste time trying to solve the wrong problem again.

  - Robin