[6bone] proposal for transfer of 6bone address management responsibilities to RIRs

[Robert Kiessling]

Bill Manning <bmanning@ISI.EDU> writes:

> % You cannot seriously recomment anyone to trust production services to
> % a network in which a single faulty BGP implementation can and does
> % repeatedly bring down half the network (as recenty seen by AS1654).
> 
> 	What does a faulty BGP implementation have to do with 
> 	a prefix?

As you perfectly know, 6bone is more than address allocation. It also
describes the network of 6bone sites connected by tunnels, with the
general philosophy of generously allowing tunnels between any two
6bone sites, and exchanging full routing tables. It's symptomatic that
the 6bone whois does not even have a syntax to describe native
connections.

In the resulting BGP topology such errors as mentioned have a rather
long lasting and global effect. We were lucky enough that AS1654's
IPv4 upstream had a 24h NOC and reacted quick to turn them off, but
you cannot take that for granted.

Or take the ghost routes as an example. It's extremely hard to debug
this problem and isolate its root cause, due to the high ratio of
interconnections, and I believe (without being able to prove it) this
highly amplifies the problem.

You could argue that this could also happen without 6bone, i.e. with
sites having RIR allocations. However, we don't see this today, and
it's unlikely that those operators will show the same attitude.

The only way I see to permanently solve these problems is to have a
core of responsible and responsive operators with sensible network
interconnections, and apply route filters to other parties connecting
to this core.

> % There's now a lot of production IPv6 connectivity available, and I'm
> % pretty sure that the backbone which now deploy IPv6 natively can agree
> % to provide sufficient IPv6 connecticity, in return for a stable
> % network.
> 
> 	Please back your assertions.

Please ask for specific connectivity requirement that you think cannot
be provided without 6bone in its current form.

What is your proposal how we can move to a stable network without the
above mentioned problems?

Robert