RE: The state of IPv6 multihoming development

[Iljitsch van Beijnum <iljitsch@muada.com>]

On Tue, 22 Oct 2002, J. Noel Chiappa wrote:

>     > if this is such a paradox, how come it has worked for years for E.164
>     > addresses?

> Well, for one, the international telephone system doesn't use dynamic
> routing.

Hm, hardly a good point. If you can train monkeys with a fax machine to
execute your routing algorithm, it must be simple enough to implement on
a router, I'd think...

> Because of the close connections between topology, addresses, and
> routing, one basically has the choice of either forcing the addressing to
> follow the connectivity topology, or forcing the connectivity topology to
> follow the addressing.

I think the last place where addressing follows topology is in old
electro-mechanical telephone exchanges, where the digit you dial is
translated to an actual crossconnect location. In IP, there is only a
slight overlap between the addressing and topology hierarchies.
Unfortunately, hierarchy -> tree structure -> no loops -> no redundancy.

Trying as hard as I can to think "out of the box" it seems reasonable to
me to see how far we can get by tying addresses to location. I know this
is an "ugly" solution as in theory, the network topology has nothing to
do with geography. And as we know, in theory there is no difference
between theory and practice, but in practice there is. Despite the fact
that I run a website under the domain "bgpexpert.com" I have no delusion
I'm the world's greatest routing expert, but on the other hand I'm not a
complete novice and I feel confident that this geographical aggregation
will work in practice. And the really good part is it's all completely
optional: you don't _have_ to geographically aggregate. If people don't
like the fact that addresses are geographically significant, they can
simply treat them as straight PI addresses and buy big routers.

> Those parts of the system that do have some number portability do it through
> a mechanism which is basically identical to IPv6 Mobility - your "phone
> number" is now a virtual 'name', and it gets translated into a different
> phone number which is your actual phone 'address'. This corresponds exactly
> to the "home address", etc of IPv6 Mobility.

It's a bit more complicated than that. Some networks use the
"intelligent network" for everything, but first you have to reach the
"home" network for a "prefix". There is a three level hierarchy here:
country, city (state in the US?), operator. In IPv4, it's just
"operator". So does this mean we're smarter or stupider than the
bell-heads?

> So there's another possible solution for people who really want to
> multi-home: use IPv6 Mobility. That's already supported in all hosts, right?

Mobility has useful mechanisms, but unfortunately it will not do
anything useful in the presence of outages.