Re: The state of IPv6 multihoming development

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

On Fri, 1 Nov 2002, Brian E Carpenter wrote:

> > Nobody is proposing a solution that needs more exchanges in order to be
> > useful.

> You may not think so, but I find it hard to believe.

What my geo proposal needs is many routers that interconnect with
routers from oether networks. Where they are located isn't important in
itself, but using geo at gives the _potential_ to have optimized
routing. Whether this potential is realized depends on the actual
topology. A different aggregation hierarchy doesn't provide the
potential for optimized routing by itself so it needs a level of
indirection to do this. This will scale better, but (unlike my geo
proposal) it can't be deployed in the very short term as it requires new
code.

> > Just to be safe I've included a statement to this effect in the
> > abstract of draft-van-beijnum-multi6-isp-int-aggr-00.txt.

> This statement will not impress business managers who see an
> opportunity to create exchanges.

Which is a good thing irrespective of the solutions adopted.

> > New exchanges
> > would only be necessary if this solution were to be used much longer
> > than intended,

> I assume that whatever gets deployed will be around for 20 or 30 years
> minimum.

There is considerable consensus here that a multi-address/locator scheme
will solve the long-term problem. This solution or set of solutions will
be around in 20 or 30 years. If my draft is implemented and not
superseded by something better in 20 years we obviously haven't been
able to agree on the requirements by then...

> > but as we approach one multihomer in 10 people even flat
> > routing for a single city such as New York, Tokio or Mexico City will be
> > problematic.

> I wouldn't expect domestic users to be MH, so I don't expect to see
> a million MH sites in those cities.

1G was the lowest figure that wasn't challenged on this list as being
too optimistic. 1G = 1:10 in ~2050.

> > Also note that as of today, NO end-user networks are multihomed in IPv6.

> Just wait until we are running real production needing 99.9% up time.

It's not the need that's lacking, but the means. The RIRs won't assign
you PI space, and the ISPs won't route it.