> From: Paul Jakma <[email protected]>
> Note that you can organise routing around AS numbers and so reduce
> routing state requirements perhaps without having to change IP
> forwarding. I.e. you can compute paths on an AS-based topology, and
> then compute IP routes based on which ASes terminate which prefixes. As
> the association of IP prefix:AS is a lot more static than the topology,
> you could reduce the amount of information that had to be passed around.
Not a bad idea, but... First, as some people have pointed out, there are
things like 'traffic engineering being done with injection of more-specifics'
to deal with. Second, fundamental characteristics of the current routing
architecture mean that deploying something like this is going to be a really
hard.
That's because the current path selection is, for better or worse, a giant
distributed computation, in which intermediate results are passed from node
to node. I.e. nobody computes the entire path; a each node does a bit, and
relies on the rest of the network to do the rest. That 'pass intermediate
results around' means that you can't really 'evolve' it to use a different
kind of data in any simple way, e.g. by a process which involves some nodes
using one type of data (addresses), and others using a different type (AS
numbers).
Even assuming you have a contiguous area of the networking using one type, and
the rest using the other, at the border between the two you have to
'translate' from one kind of data to the other. And if you have 'islands' of
stuff it gets really horrible: you have to 'translate' from one kind of data
to the other going in, and then as you exit the island, you have to translate
back.
Noel
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
