BCP for multisite multihoming

[Carlos Garcia Braschi]

2007/7/24, John Payne <john at sackheads.org>:
>
>
>
>
> On Jul 23, 2007, at 10:33 PM, Iljitsch van Beijnum
> <iljitsch at muada.com> wrote:
>
> > On 21-jul-2007, at 10:27, John Payne wrote:
> >
> >> Geo addressing sounds interesting on the surface, but every
> >> proposal seems to require a new economic model. I find it difficult
> >> to believe that will happen anytime soon.
> >
> > As the saying goes, there are many ways to skin a cat. If you hand
> > over the packets with destinations in a certain region to an entity
> > that handles that region, such as an internet exchange, then you're
> > indeed using a different economical model than we use today.
> >
> > But you can also do all the geo stuff in your own network. For
> > instance, if you have a world wide network, you could split the
> > world into 10 pieces and handle routing for each of those regions
> > only within the region. The other regions then use an aggregate to
> > get the packets to the right region. You would of course have to
> > peer with other networks within these regions or break aggregation.
> > But then, if you have a world wide network you're almost certainly
> > peering in more than 10 places as it is anyway.
>
> Yes... But I'm probably also very selective who I peer with at those
> 10 locations. Under the geo model, I have to pretty much peer with
> everyone, everywhere.
>

Not everywhere. You only need to peer in an exchange if you have local
customers in that country, and you would peer anyway in that country
to provide them with local access.

And if you limit this model for multihoming customers, you only need
to peer with other operators that do have multihoming customers... and
you can try to trust the selectivity of your multihoming customers.

Or this can be started in those IXs that do have "peer with everyone"
policies... that does not usually mean that you peer gratis. The IX
only gives out multihoming addresses to participants that have secured
peering agreements with all the others. Bear in mind that smaller
players would have problems anyway to get fair peering deals

>
> >
> >
> >> The other problem with using IXs that immediately comes to mind is
> >> that you are essentially multihoming to a single piece of
> >> infrastructure. Not really resilient in my mind :)
> >
> > Exchanges are often layer 2 networks, which aren't very resilient.
> > But money and engineering talent can fix these problems.
>
> Who's money?
>
IX costs are rather low anyway compared with international transit, so
it pays to have them resilient.