RE: Thoughts and comments

[Emmanuel Duros <Emmanuel.Duros@sophia.inria.fr>]

> From: "Walmsley, Clive" <Clive.Walmsley@dsto.defence.gov.au>
>
> CW: true and accepted, but I don't see how this changes my point. I
> assumed that a receiver will have have at least two interfaces: a UDL
> and a BDL. Further that the BDL is part of the (an) IP routed network.
> Why does the UDL standard have to consider a receiver to receiver
> scenario when the normal routing architecture on the BDL interface
> should be able to derive the route between two UDL interfaces. 

Clive,

I just want to remind that the aim of the Link layer tunneling mechanism
(lltm) is to emulate a complete bidirectional & broadcast connectivity
over a UDL. I.e., every node should be able to send broadcast and
unicast packets to every other node connected to a UDL.

Routing issues are not addresses in this document, they depend on the
type of UDL (broadcast satellite link, some unidirectional cable
modems,etc.) which may have very different characteristics (delays,
bandwidth, etc.). It also depends on policy routing based on commercial
reasons (?),... Note, that ALL sort of policy routing is possible as
long as you use the lltm. They are similar to those applied on a
'native' braodcast link such as Ehternet.

During last IETF meeting in Minneapolis we had several presentations on
what experiments have been performed on UDL using Mcast routing
protocols. The working group is expected to write a new draft describing
these experiments.

The minutes of the last IETF meeting should be sent soon on the ml and
will give pointers to the various slides. You may already have a look
at:
http://www-sop.inria.fr/rodeo/personnel/eduros/various/udl-example.ps.gz
http://www-sop.inria.fr/rodeo/personnel/eduros/various/udl-experiments.ps.gz


> Note that I have assumed that a receiver would not distribute routing
> information stating that had access to a route to its own UDL interface
> or traffic arriving on that interface. Further that the send only feed
> would distribute on its BDL interface the fact that it had routes to all
> the receivers UDL and BDL addresses. I am not certain how these would be
> derived at present by the send only feed. 

This depends on the routing protocol you are using. If it is DVMRP, a
receiver must propagate on the BDL the routes learned from a send only
feed (sof). Downstreams routers must be able to calculate the reverse
path multicast tree up to a source via a sof in order to forward a
multicast datagram which came from a UDL. If it is RIP, you may not want
to take into account the routes advertized by a sof. If you do, the
receiver will forward unicast traffic to a destination via a sof,
i.e. the traffic will be encapsulated in the back-channel of the
lltm. It is probably not the right thing to do...

Regards
Emmanuel Duros