Re: Ideas for IPv6 BGP and tunnelling

[Adrian Kennard <[e-mail]>]

Ole Troan wrote:
>> But I fear there so far the idea hasn't gotten much traction. In fact, the
>> RFC4798 predecessor documents [1] included ability to set up tunnels over
>> GRE and similar non-MPLS encapsulations.  This was explicitly _removed_
>> because the solution was targeted at MPLS networks, not as a general purpose
>> BGP-signalled tunneling mechanism.
>>
>> [1] take a look at e.g:
>> http://tools.ietf.org/html/draft-ooms-v6ops-bgp-tunnel-00
> 
> I thought 6PE and BGP tunnelling got split into separate documents?
> obviously my memory isn't serving me right.
> 
> you can still do BGP tunnelling with existing mechanisms. PEs are
> connected through a full mesh of BGP peerings. each PE has an
> automatic tunnelling interface (6to4, automatic tunnelling). BGP
> next-hops are the 6to4/v4compatible address. note that 6to4 is only
> used internally and the sites connecting to the PE uses native
> addresses.

OK, it is specifically the use of the next hop that I wanted to check. 
If that is covered by an existing RFC then great.

The idea is the use of a suitable IPv6 next hop for an IPv6 prefix 
announced via an IPv4 BGP session. Is the correct practice to use a 
2002:: prefix address as the IPv6 next hop to make the routing send via 
a protocol 41 IPv4 tunnel then? Is that documented?

There is another more radical idea I also have which maybe I'll raise as 
a separate issue though :-)