The issue is not just how many routes any box (or set of them) can hold
(although that is a significant concern, because there is installed base all
over the world, and not all the component organizations have budget to
replace/upgrade it all freely).
I am not suggesting massive worldwide upgrade :) I am convinced that
currently deployed routers in the Internet core would do just fine with
the help of just control plane changes. Contrary currently proposed
schemes do require hardware changes/additions for interdomain tunneling.
The thing is that today's routing architecture is a giant distributed
computation, one which doesn't 'run to completion', but is always running,
responding to changes in the world-wide physical connectivity. A change
happens, and things respond, and eventually affected paths settle down and
stabilize after some time Ts.
That is true and the possible upcoming impact of knee effect of IPv4
addresses should not be dismissed. But my observation lead me to believe
that various proposals here are building a Titanic solutions where
what is needed to be on the safer side is just a new operational mode to
the current boat.
In particular to increase overall system stability and Internet safety
margin I would recommend to consider a very simple approach ...
To separate in current today's BGP prefixes from next hop. To converge
next hops aggressively while increasing the timers for prefixes. Very
simple and IMHO quite effective way to reduce the product of "# of
prefixes * change rate".
And such separation could be just new SAFI for next hops. Easy to
gradually deploy, does not require new mapping layers, can be user with
intra or inter domain tunneling (but this is not a must) ....
To summarize I am not convinced .. why we are jumping into the ocean
while there is plenty of room in the current pool to swim.
rtgwg mailing list