IPv6 teredo blackout June 8th

[Martin Millnert]

On Wed, 2011-02-02 at 21:15 +0100, Daniel Roesen wrote:
> Question is: what do we have in the toolbox currently? 
> 

Well, a DIY Linux with a up-to-date Intel CPU and sufficient mobo PCI-E
routing will do quite a few packets; without question fill a GE pipe,
and I believe reach quite high on a 10G nic as well (multiqueue support
in 82599 together with multicore CPU:s goes a long way to overcome a
vanilla Linux IP stacks shortcomings when it comes to forwarding
performance).

Cisco does it in HW, people say.

> Also we should think of a BCP how to deploy such relays (I never
> operated a globally advertised anycast 6to4 relay myself, so cannot 
> comment or guesstimate how many things we can do wrong when trying to
> do so).

A draft is being prepared for this purpose and shouldn't be very long
before it's finished.

Ole, unless you completely wipe the DFZ of the relay prefix, you will
still be seeing the few remaining massively overloaded relays forwarding
really sucky packets to the web servers.  And even if you do rid it from
the DFZ, you still have to also de-configure it on all hosts...
Although an entertaining idea, it doesn't really strike me as very
feasible.

Cheers,
-- 
Martin Millnert <martin at millnert.se>