Some weeks ago Robin asked whether my hierarchical routing model would be
incrementally deployable
and I responded by saying "No", without having given any thoughts to this
aspect. But in the meantime I did
and my answer to his question is "Yes, for sure".
Hierarchy means logarithmical reduction of the complexity.
E.g. instead of 50 x 50 x 50 x 50 (which is 6,250,000) you need only 50 + 50
+ 50 +50 = 200.
Whenever I said instead of 200-300K FIB entries, 2-3K entries will be
sufficient, I even might have been too cautious. The above figures mean: a
BGP-FIB
with up to 6,250,000 entries can be replaced by 200 entries.
However, 200 is such a neglectable number, that it should be affordable IN
ADDITION to the current BGP-FIB figure as well. Consequently for the next hop
determination towards a far remote destination a router may have two
alternatives, and as I said before, for the last few hops towards a
destination in
the neighboring network part, the existing method could be applied (even for a
longer while).
Also: this little new forwarding information can generously be placed into a
bigger-sized table such that the next hop towards any genuine hierarchical
node can be done by ONE SINGLE matrix element look up ! (no search for a
fitting match).
Also: Eliminating the scalability problem is not all. Indeed, this little
but well aquired new routing information can do much more than BGP or LISP:
Note that Multihoming is only focused on multiple alternatives with respect to
the last hop. The right focus however would be multiple alternative paths
whose last hops involve alternative ISPs.
Heiner
