RE: The state of IPv6 multihoming development

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

    > From: Iljitsch van Beijnum <iljitsch@muada.com>

    >> The benefit of multi-homing has a price - the only question is, who's
    >> going to pay the price? Simple fairness demands that it *ought* to be
    >> the entity which directly benefits.

    > So who benefits if Google is multihomed? The user or the server? Or
    > both?

If we only had to support multi-homing for Google, CNN and the N other top
sites (for a reasonably small N, say 1K or so) we'd be done.

The problem comes when you want to multi-home many, many small sites, all of
whom (by definition) will be conversing with only a limited subset of the
network. Forcing the rest of the network to help pay for the benefit they
receive from multi-homing is patently unfair.

I do agree completely with the basic idea behind one observation in Tony's
draft: "the overriding assumption that the Internet will have to be built
using one or the other rather than leveraging the strengths of each in
context." There's probably no single mechanism which is "the" answer to
multi-homing; you probably need several, each with its own optimal
applicability domain.


    > In a multiple address solution this is much, much worse: in that case,
    > all IPv6 hosts may have to implement extra functionality to be able to
    > talk to multihomed destinations.

How much extra code is it, really? It can't be as much code as IPv6 Mobility,
or Authrntication, each of which is already mandated.

In any event, if you think it's too much extra complexity, use IPv6 Mobility
instead. That way you'll also get connection failover if you have to switch
addresses dynamically because one of your incoming links went down.


    > But we're not debating straight PI, but geographically aggregatable PI
    > (GAPI).

All forms of PI have the same basic problem, laid out in the previous
messages, which is that they are ""topological-location-independent
topological-location-names". So which particular flavor of PI it is really
isn't of any interest.

    > Only when the routing tables grow beyond that, networks in the affected
    > region need to install more memory in existing routers or more routers.

You have a static view of a dynamic reality. There are a lot more factors
than just memory size.


    > the real question is whether multihoming in IPv6 will be good enough to
    > get people who multihome "for free" to move to v6.

Hmm. If the current multi-homing "solution" (where the load of supporting
multi-homing is borne entirely by the routing) "works" (and I don't think it
does, but let me play advocatus diaboli here for a second), why not adopt it
"as is" for IPv6?

Contrariwise, if it doesn't work, why are you trying to adopt a variant of it,
a variant which is so little different in its underlying technical
implications that it amounts to the classic "re-arranging the deck-chairs on
the Titanic"?

	Noel