RE: draft-kurtis-multihoming-longprefix comments

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

On Tue, 7 Jan 2003, J. Noel Chiappa wrote:

>     > Today, if routers are owned by the same owner, they hold the same
>     > routing information. Maybe it's time we try aggregating based on
>     > something else than router ownership.

> Reading this, I literally had to cringe and hold my head in my hands.

:-)

> The *only* way to aggregate *which is useful to the routing* is based on
> *actual connectivity* - i.e. the actual ability send packets. Ownership of box
> X and box Y is basically irrelevent (although policty routing may use it for
> *additional* constraints), from the routing's point of view. If box X and box
> Y are connected by a wire - now, that's different.

Suppose ASes 1, 2, 3 and 4 are all present with one router in Chicago,
NYC, Dallas and LA and they all interconnect there. Each router has 100
customer routes. So you can aggregate and have 400 internal and 3
external routes in each AS. Or you can do geo aggregation as per my
draft and have 400 LA routes regardless of which ISP each route belongs
to in LA, 400 Chicago routes in Chicago and so on and 3 pilot routes to
other locations. So it's exactly the same thing, only in this case you
can aggregate multihomers.

Now obviously in the real world interconnection between regions within
an ISP network is denser than between ISP networks within regions, but
regional interconnection is still dense enough to make it worth the
trouble.

But in the mean time I'm working on a multiple address draft which is of
course much cooler but upgrading all hosts in the world also has its
problems...

Iljitsch