In the last post in this series, I described several types of providers — and even how those descriptions are no longer really “pure,” for the most part (although NTT, for instance, is a pure transit provider that only offers a few services throughout the world). For each piece of a provider’s business, then — transit, upstream, and settlement free — how does the provider offer value that they can charge for? The answer lies mostly in controlling traffic flow between the edges of the network to maximize revenue while minimizing costs. Let’s look at three examples to get a feel for how this works.
Assume AS64500 and AS64503 are both “customers” to begin with, and AS64501 and AS64502 are providing transit services for a fee. Where does AS64501 have the most potential to create a revenue stream? The primary connections on which AS64501 can collect revenue are along [A,C] and [K,J]. How does AS64501 make money on these links? Because their customers — AS64500 and AS64503 — pay based on the size of the connection (most often) or the amount of traffic transited over these links. How, then, can you maximize revenue on these links? The simple answer is by increasing the traffic flow across them. Where does AS64501 spend the most money? Along the red “geographical separation” line, which falls on the [F,G] link. So the point of AS64501’s policies is going to be maximize traffic on paying links, and minimize traffic on non-paying links, particularly links that cost a lot of money to run. How does this play out in terms of policy?
First Example: Customer to Customer
Assume traffic is coming in from A and is destined to K. Since AS64500 is a customer, the operator will prefer to make money on both ends of the traffic stream, and hence the policies will be configured to carry the traffic along the path [A,C,F,G,J,K] — all the hops will be within the provider’s network. To accomplish this, the provider will (generally) set their BGP local preference so that routes destined to a directly connected customer are preferred over routes through a peering provider. This is what we might call “cold potato routing,” or rather keeping the traffic within your network, although you’re pulling traffic across an expensive link, in order to maximize revenue.
There is another aspect to this, as well — customer experience. If the customer calls the AS64500 NOC, the provider wants to be able to see the entire traffic flow, end-to-end, if possible. In order to control the customer experience, you have to control the traffic end-to-end.
Continued next week.