Re: The state of IPv6 multihoming development

["J. Noel Chiappa" <jnc@ginger.lcs.mit.edu>]

    > From: "Michel Py" <michel@arneill-py.sacramento.ca.us>

    >> If you have a host which is multi-homed to widely spaced
    >> points in the topology, there's unlikely to be a routing-based
    >> solution, i.e. one where that host has only one address.

    > I totally disagree

I'd be interested to hear what your reason is for thinking that this
statement is incorrect.

    > Would you be more specific about why you find this unlikely, and
    > what you call "widely spaced" (which I read as "from different PA
    > blocks).

Addresses are a symptom of the problem, not the cause.

By "widely spaced ... in the [connectivity] topology", I mean just that: the
host H is connected to the Internet in two places, which are far apart in the
connectivity graph. I.e. the shortest path between the two connection points
(not through the host) is long.
E.g. suppose this host is connectes to two local ISP's L1 and L2. Further
suppose these are connected to upstream larger ISP's U1 and U2
(respectively). Further suppose that U1 and U2 don't peer near the points
where the local ISP's connect to them. All hardly unlikely, but you've
probably got at a 15-hop path, or something.

No matter *what* kind of address you give the host (be it connectivity-based,
or "provider-independent"), if one of the interfaces is down, the routing
table at the midpoint of the path (i.e. where incoming traffic is going to
have to decide which way to go, "left" or "right") is going to have to have a
currently correct entry *for this specific host* if the traffic is going to
get there. In other words, you have to have host routes, propogated over a
wide scope (back to a major top-level ISP peering point, in the example
above). This is not feasible.

That's why I say "addresses are a symptom of the problem, not the cause",
because the problem is the same no matter what kind of addressing scheme you
are using.


    > There is a difference between the solution itself using multiple
    > addresses and each host using multiple addresses. If each host has to
    > use multiple addresses, I regret to report that Craig's statement is
    > not an opinion, but a fact. Any people that disagree with this and that
    > actually have renumbered a 1000+ subnet enterprise network, please
    > speak up

Who's talking about (manually, I assume) renumbering a 1000+ subnet network?
IPv6 hosts get the high bits of the globally-unique addresses from their
routers, right? Why is it so hard/insecure to get two (or more) instead of
one?

	Noel