[e2e] node addresses vs. interface addresses

[J. Noel Chiappa]

    > From: Joe Touch <touch at ISI.EDU>

    >> it's putting the entire cost of multi-homing .. into the routing,
    >> where the cost is paid by everyone (effectively - long arcane routing
    >> point elided) across the network

    > Except again that this can be hidden from the rest of the network using
    > tunnels. A host can have a single endpoint address - on an "internal"
    > virtual interface, used through a set of tunnels, each using outer
    > addresses based on one of multiple real interfaces.

I must be missing something here. (I looked at the ICNP paper, but it had
nothing about tunnels, and the tunneling I-D didn't have much to say about
multi-homed hosts.)

Although you didn't provide full details of how it works, when I try to think
of schemes using tunnels to do multi-homing, either i) the host on the other
end emits wrapped packets (i.e. with the outermost destination IP address
being one of the "multiple real interfaces"), in which case this isn't a
general solution, since not all correspondent hosts will be prepared to do
that, or ii) you doing something funky, like either a) NAT, or b) using
something that looks like IPv4 mobility, with a "base-station" kind of
address?

I suspect that you mean the latter; the packet is then forwarded from the
"home station" (as I'll call it) toward the real destination through a
tunnel, wrapped in a packet with the address of a real interface on the
destination host. (In fact, in some variants I can think of, in normal
operation the host can be its own home station.)

If that's what you're doing, you haven't really changed anything. Here's why.


The home station address represents a renewed point of vulnerability. Even if
there are multiple paths from the home station to the multi-homed host, if
the connection to the home station from the rest of the Internet is not
redundant, service to the end hosts is still at risk. So you haven't
eliminated the need for multi-homing, and thus the multi-homing problem, just
moved it one step away.

Now let's think about how to provide redundancy to the home station. To start
with, it has to have multiple links. How do people know which links are up,
and how do they get to them? Through the routing, of course.

If the home station is a large service that's shared among many, many users,
then we can afford to advertise *its* address across the network through the
routing. But this effectively is almost exactly that what we can have now,
where a large number of customers sign up with ISP X (which has multiple peers
for redundancy, and because of it size can economically be globally routed) -
provided that any who want to be multi-homed get multiple drops from that ISP.
They don't even need multiple addresses - the intra-ISP routing can pick an
inbound link which is up.

(Arcane point: your scheme allows the multi-homed hosts to get the multiple
services from different ISP's, which is perhaps useful.)

If the home station is not shared among large numbers of users, then we can't
afford to advertise a route to it globally.

In other words, on a deeper look, the situation is indeed exactly what the
initial comment above indicated: "you haven't eliminated .. the multi-homing
problem, just moved it one step away".

Am I missing something?

	Noel