At 02:30 PM 5/5/2003 -0400, bdragon@gweep.net wrote:
What about those who are affected by state generated by someone else's customers? Providers are already compensated by their customers.
To a first approximation, peers have business relationships. How money flows (if any), how much bandwidth is available between the peers, how far the peers will pass traffic for each other---those are all parts of the business relationship between the parties. I believe that a fourth factor--the amount of state that each are willing to accept and carry for the other--should be added to the equation.
The issue is when Joe Bob ISP's customer deaggregates their /16 into /24s for their own traffic engineering purposes which the rest of the internet has to bear without cost recovery.
Cost recovery would be nice. The ability to make a profit would be even better. The goal should be to financially reward the parties that can handle complex state in their networks. We also want to financially reward parties that do route aggregation. Providers should have a financial incentive to hand out provider-managed space to their customers. The obvious incentive is for the provider to advertise a small number (one?) of short prefixes, with the customers getting longer sub-allocations. That's how providers might recover the costs of managing their provider-managed spaces.
The problem is how these 3rd parties bill Joe Bob's customer for their deaggregates, and how to identify approved (paid for) deags vs. unapproved deags.
As you say--organizations de-aggregate to engineer where traffic flows. Why are they motivated to do this? Because the costs of circuits are a known quantity, and the cost of de-aggregation roughly zero. So a fiscally rational organization will de-aggregate whenever and whereever possible. If, on the other hand, each prefix advertisement had a cost associated with it, then an organization would be able to make a de-aggregation decision based on its local requirements. The organization might even decide not to de-aggregate at all, if the prefix advertisement costs were high enough to justify additional circuit costs. Consider the possibility of charging a peer one rate for transit state, and a lesser rate for no-advertise state. That's what we do today already to control bandwidth costs.
The high-level model is fairly easy: 1) largest aggregates are free 2) more-specifics from within the same originating ASN are charged at rate X which is adjusted proportionally to prefix length (as length gets longer, the rate goes up). 3) more-specifics from within different originating ASN are charged at rate Y (Y<X since meaningful content is more likely to exist), and again, adjusted proportionately to prefix length.
Why should advertisements of varying lengths cost more (or less) than any other? It's the routing table entry that costs, not the length.
It is only when you get to the specifics that things start to break down. 1) The prior identification of permitted/rogue routes 2) The means through which these rates get set 3) The means through which an ASN can contact 3rd parties in order to provide payment. 4) The means through which an ASN can verify that the 3rd party has accepted their route, or is even eligible for payment (someone not running BGP isn't having any resources consumed).
Since this sort of settlement basis is (in my opinion) doomed to fail, the only other approach is what we have now, filtering everyone everywhere. What can and should be done is that the line in the sand is determined, agreed to, and adopted by everyone.
Let me be clear--I don't believe that settlement should happen anywhere other than within existing peering relationships. Recall that I'm also proposing that netblocks be treated as property, able to be bought and sold. Part of the transaction cost would involve paying a registry to register ownership, just as real estate or trademarks are registered today. Clearly identified ownership of netblocks goes a long way to meeting your Issue #1. Issues 2-4 (in my opinion) shouldn't be set industry-wide, because circumstances vary widely on a temporal and spatial basis. I believe it's better to structure the financial incentives than to impose such rules. === Bill Nickless http://www.mcs.anl.gov/people/nickless +1 630 252 7390 PGP:0E 0F 16 80 C5 B1 69 52 E1 44 1A A5 0E 1B 74 F7 nickless@mcs.anl.gov