Need to develop mechanism (feedback) to discourage local behavior that is non-cooperative from a global routing and addressing point-of-view.
Improving routing information aggregation and increasing address space utiliaztion depends to a large degree on subscribers:
Improving routing aggregation depends on providers
Propagate each route to as few domains as possible, while maintaining connectivity among as many domains as possible.
If I use a CIDR route and move from provider A to provider B, what happens when I move to provider A advertisement? Provider A does not have change their CIDR advertisements.
How big is small and medium? How much money do you have? The more money that you have the less frequently you will have to renumber.
In the multi-home case, how far up stream does the subscriber play? What is the address allocation policy? If the addresses are not aggregatable, the subscriber has to pay to push them to the default-free zone of the Internet.
Will charging IPv4 addresses accelerate the desire for IPv6 addresses? Maybe. We don't know yet. Microsoft has shipped a bunch of DHCP systems that can be renumbered (fairly) easily.
What happens if the providers that charge for routes vary the charge in interesting (and perhaps market driven ways)? I did not address this in this talk. Whatever folks charge should be publically available.
Do we really have to do this? Does the user need to see this? It should be presented as a discount, not an incremental cost.
Milo Medin says:
The issue is more than just router capacity. There is also operations support and managing issues. IPv6 does not really solve this problem. Automatic renumbering in both IPv4 and IPv6 really does work similiarly and therefore cost about the same. NAT boxes work quite well. This should be considered more heavily. If the charging mechanism is implemented, it will be bilateral and charging will probably not be fair.
Yakov says:
There are greedy IPS and incompentent ISPs. That will certainly motivate some of this. NAT boxes are good.
The recent change at the NIC on charging for the domain name seems to work. Should the route charging mechanism be similiarly done? It seems that passing the charging up and down the routing system is cumbersome. Of course, there is a problem of distribuing the back will be hard. Yes, it probably is cumbersome. It will need to be govened by the contract and bilateral agreements.
How would route pulling really work? A string of bilateral agreements would need to be inplace to provide for route pulling.
Why not charge based on prefix length? You could, but it's not required to improve route aggregation.
How would a multi-homed customer be affected? Customer would pay to push unaggregrated routes to default-free Internet. The Customer may also want to play for pull to cause routes to be predictably available.
Sean Doran says:
The real problem is getting more capabilities into the routers. This costs money and this money must be acquired from somewhere. There are still a fixed number of routing slots [in the Cisco Routers that Sprinlink uses]. Since there is a fixed number of routing slots, there is only so much that can be done before the cost becomes infinite since the slots are full.More aggregation makes more slots available and therefore aggregation is still the best approach to keeping costs down.
If more slots are going to be used, then someone will have to pay and that will have to be negociated in various bilateral agreements.