IPv6 PI allocation
Iljitsch van Beijnum
iljitsch at muada.com
Fri May 18 14:54:26 CEST 2007
On 17-mei-2007, at 17:10, David Conrad wrote:
> Do we have a ROUTING TABLE PROBLEM now? Nope. Don't worry, be happy.
Strange. I distinctly remember spending two days in Amsterdam last
fall with with a group of people that included half the IESG and IAB
and builders and users for the largest routers, and the consensus at
the end of that meeting was that there is a problem, and that it's
In my experience, when some people say there is a problem X and some
other people say there is no problem X, there IS a problem, although
it's not necessarily X.
Unfortunately, the decision making process in our industry is
fundamentally flawed. Within the IETF, it's only possible to do
something if the vast majority agrees about the action, with the
result that many necessary actions are never undertaken. Address
allocation policies have global impact, but are created in five
regions more or less independently, by people who mostly represent
themselves and their own particular interests. This way, you can
easily end up working on a mechanism for scalable multihoming
(shim6), and then have non-scalable multihoming (PI) introduced just
as the scalable mechanism is defined.
The interesting thing with global routing is that it can't fail.
Since any change is visible throughout the network within one or two
minutes, anything that breaks global routing will be rolled back
immediately, rather that be allowed to break what we have today.
Obviously using a 15 year old routing protocol that can be
characterized as "RIP on steroids" and the insistance that all
routing information injected into the global table anywhere in the
world must be visible throughout the entire network is going to have
bad consequences in the long run.
Let me know when the I* community is ready for something better.
More information about the ipv6-ops