BCP for multisite multihoming

[Iljitsch van Beijnum]

On 27-mei-2007, at 0:10, Simon Leinen wrote:

> ISPs generally feel strongly about their rights to decide for
> themselves with which other ISPs they want to peer (settlement-free).

And rightly so.

> Proponents of new schemes should make it very clear that they won't
> take this right away from ISPs, if they want buy-in from that
> community.  Personally I think geographical addressing could lead to
> better aggregation, even without imposing full interconnection
> (settlement-free or not :-) within a geographic region.

If you only peer in one location for traffic towards a certain  
geopgraphic area, then routing is simple and very scalable: in  
theory, only one router in that specific location has to know all the  
specific routes for that geographic area; all other routers only see  
an aggregatte.

Today, all default-free routers must know all the specific routes for  
the entire world.

The first extreme is obviously bad for business for ISPs (not to  
mention fragile, what if your single location becomes unreachable?)  
and the other extreme has scalability issues (as a matter of  
principe--whether it keeps working in practice is a different  
question). Some kind of knob that allows an ISP to move from one  
extreme to the other would allow the level of geographic aggregation  
to be traded off against investments in more or larger routers or  
less optimal traffic flow, making it a business problem for ISPs  
rather than an engineering problem for router vendors.