RE: I-D ACTION:draft-huitema-v6ops-teredo-00.txt

["Christian Huitema" <huitema@windows.microsoft.com>]

> One thing to note here: in addition to "Server", the Teredo
architecture
> also includes "Relay".  And while only a small fraction of packets go
> through the server (AFAIR), a large number of them could go through
Relay
> (which are probably often one and the same).

The traffic between a Teredo host and a native IPv6 host will go through
the relay closest to the native IPv6 host. This relay does not have to
be the same as the Teredo server. In fact, in our beta deployment, no
relay traffic whatsoever goes through the server. In the beta
deployment, the native IPv6 hosts are dual stack, and they use their
IPv4 connection to send UDP packets directly to the Teredo hosts, using
what we call a "host specific relay".

Now, there are cases where native hosts cannot run a "host specific
relay". Maybe they don't have adequate IPv4 connectivity, e.g. because
they are located behind a firewall, or maybe their stack cannot be
upgraded to include host specific relay. In these cases, the IPv6
traffic to Teredo hosts will be sent to the closest IPv6 router that
advertises a route to the Teredo prefix (3FFE:831F::/32 in the
experimental deployment). The traffic will be carried over UDP between
the Teredo host and that router, in both directions.

> So, in practical purposes I believe Eric's comment about "server" can
be
> generalized to be "server or relay".

The comment was about reliability of the transmission. Reliability is
assured end-to-end, e.g. using TCP.

> In that vein, please remember that people MAY NOT deploy Teredo as
some
> may envision it, to be used mainly between Teredo hosts only -- but
> between the hosts and *the rest of the IPv6 Internet*.

There may be some misunderstanding here. Even in the beta deployment of
the Three Degrees application, at least half the hosts are getting IPv6
connectivity using 6to4 or native services. The desired outcome is the
deployment of a large number of Teredo relays at locations close to the
native IPv6 users, to avoid dog leg routing through remote servers. 

I expect each of these relays to serve only a small community, i.e. the
networks to which they advertise the route to the Teredo prefix. In the
Teredo design, a relay only has to accept Teredo traffic from the IPv4
addresses to which it has recently sent Teredo traffic; the relay could
in fact implement a form of stateful filtering.  

> If you want to deploy IPv6 Internet and IPv6 Teredo Internet
separately,
> you're certainly correct: few packets pass through servers/relays.
> 
> I do not have that deployment scenario in mind; quite the opposite
:-).

Neither do I.

-- Christian Huitema