Re: Teredo spec has been published!

My reason to raise such a topic is the concern that current Teredo specification may cause routing problem in the real ISP network, because of the fact that many ISPs wanna control the routing policy of their own IPv6 network. And this problem may cause ISPs unwilling to deploy Teredo service.

1. Based on current Teredo spec, the core router in an IPv6 network shall advertise 2001:0000::/32 route through BGP4+ to other ASes, to ensure the connectivity of its own Teredo users. Also, I think it's better for the core router to also act as a Teredo relay router, otherwise too many static routes have to be configured. But how can we expect the IPv6 core router to have the native IPv4 connection?

2. In some situation, the traffic from a IPv6 server to a Teredo user in ISP network A may bypass ISP B's network, if the Teredo relay router of ISP B is nearer to the IPv6 server. So for all network operators, it's hard to predict and control both the IPv4 routing path and IPv6 routing path.

Vicent gave the good point that 2001:0000::/32 should be the default value, as recommended by the IANA, but I recommend that this prefix be configure-able in Teredo server. In such case, the ISP provides the Teredo server, and the ISP controls IPv6 routing policy.

2006/2/6, Christian Huitema <huitema@windows.microsoft.com>:

> Off a slightly different tangent: what sort of routes are expected to
> show up from 2001:0000::/32 in the routing table?

I would expect a single 2001:0000::/32 route leading to the closest
teredo relay.

> I didn't read recent Teredo drafts, but if I remember the concept
clearly
> enough, the embedded server (relay?) address would make this "one /64
> prefix per server", right?

Only a tiny fraction of the traffic goes through the servers. The bulk
of it goes on a direct IPv4 path between the relay and the Teredo host.
The benefits of "per server' /64 routes would be marginal and the impact
on the routing tables would be large.

> Or will all relays announce the whole /32, effectively anycasting it,
> as it is done with 6to4 2002::/32?

Yes. With this design, traffic will go from IPv6 host to the closest
Teredo relay and from there on a direct IPv4 path to the Teredo host.
This is expected to be a good approximation of the shortest path between
the IPv6 host and the Teredo host.

-- Christian Huitema