Re: Regionally aggregatable address space for multihoming

[Brian E Carpenter <brian@hursley.ibm.com>]

How would this work for intercontinental private networks,
whose multiple connection points are in several different
continents?

   Brian

Iljitsch van Beijnum wrote:
> 
> Hi.
> 
> I guess I'll jump right in.
> 
> After reading "IAB/IESG Recommendations on IPv6 Address Allocations to
> Sites" it occurred to me that a way to keep the number of routes in the
> routing tables down could be to use regionally aggregatable address space for
> multihoming.
> 
> The idea is that routers on one end of the globe have little need to know
> very specific routes to multihomed networks in some another part of the
> world. For instance, a router on the US west coast doesn't need to know how a
> network in Europe is connected. It should just send the packet to the
> east. At some point, the packet would end up in the destination region, or as
> close to the destination region as the originating network extends, and only
> there the originating network would have to know where to send the packet
> next.
> 
> Individual networks could implement filters in strategic places to keep the
> specific routes for multihomed networks inside the proper region. There is no
> requirement that networks use the same regional subdivision: a European
> network may use regions that map to two countries, while another network that
> has only a few points of presence in Europe may see the entire continent as a
> region. Some networks may prefer to keep carrying all routes everywhere.
> 
> All networks announce their customer's routes everywhere, but peers simply
> filter out these announcements on exchange points that do not fall within
> their idea of the region. So our European network would normally only accept
> routes to Dutch addresses in Amsterdam and Brussels (for backup) but the
> other network only connects to the exchanges in London and Paris so it would
> pick up the announcements for the Dutch multihomed address range there.
> 
> A further reduction of the routing tables could be possible by installing
> exchange routers for regions at internet exchanges. The exchange router would
> announce an aggregate for the region and all networks may send traffic to
> that router. The exchange router holds all the more specific routes for the
> aggregate and forwards the packets to one of the providers for the
> destination network.
> 
> A problem with this is that all networks that have multihomed customers in a
> region must connect to all the exchanges where the aggregate is announced.
> All other default free networks have to connect to at least two of the
> exchanges for that region. Exchange routers should be present outside as well
> as within the region they service, for reasons of redundancy and because not
> all networks are present in all regions.
> 
> The granularity of the address assignments has to be fairly fine, so that it
> doesn't conflict with the hierarchical structure of any existing network. A
> granularity of 1 to 10 million people to a region seems reasonable. (This
> would mean large cities or small countries/states/provinces.) With
> potentially 1% of the population being multihomed that would make for about
> 100k routes per region, so today's routers would be able to handle several of
> those regions.
> 
> Iljitsch van Beijnum