Re: [udlr] Questions

[Steve Deering <deering@parc.xerox.com>]

Christian,

> 1) point to point packets are sent encrypted with the intended recipient's
> key -- that key being the DVB equivalent of a media address, obtained with
> the DVB equivalent of ARP. Hiding a small home network behind a single
> antenna would require the equivalent of proxy-ARP, that is declaring
> all authorized receivers in some central database.

OK, I see the problem.  I don't suppose you'd consider the alternative of
sending the traffic in the clear, and relying on IP end-to-end encryption
(e.g., IPSEC ESP) for privacy?  Even then, you'd have to do something
non-standard at the receiving routers, i.e., have them accept or ignore
incoming packets based on what IP subnets are behind them, since there
would be no link-layer information that would identify the corect receiving
router.  Also, imposing the notion of "behindness" is undesirable.  Bad
idea -- forget I suggested it.

> 2) radio reception sometimes fades away. If you use stritly static
> routing, you end up with a black hole. Individual holes can cure this by
> being multi-homed with a satellite plus a normal address.

Here's another kludge: if the router attached to a satellite receiver
decides that satellite reception is unacceptable, it can modify the metrics
of the subnets that it advertises over the terrestrial link to make the
terrestrial link be the favored path to those subnets.   Lots of problems
with that idea too.

> You may imagine that all the hosts of a small home network are also
> multi-homed

Are you using "multi-homed" to mean multiple physical interfaces or just
multiple IP addresses?  Was your point that the alternative addresses
ought to be made known to the satellite transmitter routers, even though
those addresses don't belong to the prefix allocated to the satellite net?
This is the standard question of whether to leak addresses from one
provider into another provider's cloud.  If that's done on a large scale,
routing tables explode.

> Multi-homing and proxy ARP scale as 0(N), with N the number of hosts in
> the "local network" or "home network". For some value of N, you are better
> off doing the right thing, i.e., explicit routing using BGP or RIP, or
> DVMRP for MBONE feeds.

Yeah, the issue I identified (of millions of neighbors) is simply the
large-cloud issue, for which there are no good answers known.  The best
we can do, apparently, is divide the large cloud into multiple logical
IP subnets, with different LISes being fed from different feed routers
(possibly sharing a single satellite transmitter).

Steve