The authors describe how Internet is operated by thousands of interconnected Internet Service Providers (ISPs), with each ISP interested in maximizing its own profit. Although ISP cooperation is necessary in order to provide Internet services, without an appropriate profit-sharing mechanism, profit-seeking objectives often induce various ‘selfish behaviors’ in routing and peering interconnects, degenerating the performance of the network and the services it provides to the end user. This issue is at the heart of the NetNeutrality debate, and shows the need for an appropriate compensation structure that satisfies all types of ISPs.
One example of this selfish behavior is when US’s Level 3 unilaterally terminated its settlement free peering relationship with Cogent in 2005. This disruption resulted in at least 15% of the Internet being unreachable for the users who utilized either Level 3 or Cogent for Internet access. Although both companies restored peering connections several days later with a new ongoing negotiation, Level 3’s move against Cogent exhibited an escalation of the tension, and showed that new forms of ISP settlements are needed.
Compared to the traditional settlement models in telecommunications, the Internet has exhibited a more versatile and dynamic structure. The most prevalent settlements a decade ago were in the form of bilateral negotiations, with both parties creating either a customer/provider or a zero-dollar peering relationship. Today, because of the heterogeneity in ISPs, simple peering agreements are not always satisfactory to all parties involved, and paid peering has naturally emerged as the preferred form of settlement among the heterogeneous ISPs. Nevertheless, the questions like “which ISP should pay which ISP?” and “how much should ISPs pay each other?” are still unsolved. These open questions are also closely related to the NetNeutrality debate, which argues the appropriateness of providing service and/or price differentiations in the Internet.
Borrowing from the well-known economic concept – used in coalition games - called the Shapley value, and applying it to a general network setting, the paper proves that if profits were shared as prescribed by the Shapley value mechanism, not only would the set of desirable properties inherent to the Shapley solution exist, but also that the selfish behaviors of the ISPs would yield globally optimal routing and interconnecting decisions. These results demonstrate the viability of a Shapley value mechanism under the ISP profit-sharing context and the implications on the stability of prevalent bilateral settlements and the pricing structure for differentiated services in the Internet.
The authors look at the Internet as containing three classes of ISPs:
Eyeball ISPs, such as Time Warner Cable and Comcast specialize in delivery to hundreds of thousands of residential users, i.e., supporting the last-mile connectivity.
Content ISPs specialize in providing hosting and network access for end-users and commercial companies that offer content, such as Google, and Yahoo. Typical examples are content distribution networks (CDNs).
- Transit ISPs. Transit ISPs model the Tier-1 ISPs, such as Level 3, Qwest, and Global Crossing, which provide transit services for other ISPs and naturally form a full-mesh topology to provide the universal accessibility of the Internet.
The key results cover the following
aggregate revenue can be decomposed by content-side and eyeball-side components; and
costs can be decomposed with respect to individual ISPs. Each revenue/cost component can be distributed as a Shapley value of a canonical subsystem to each ISP in the coalition.
- aggregate revenue can be decomposed by content-side and eyeball-side components; and
The Shapley value solution justifies:
why the zero-dollar peering and customer/provider bilateral agreements could be stable in the early stage of the Internet;
why, besides operational reasons, paid-peering has emerged; and
why an unconventional reverse customer/provider relationship should exist in order for the bilateral agreements to be stable.
- why the zero-dollar peering and customer/provider bilateral agreements could be stable in the early stage of the Internet;
- Instead of supporting or disproving service differentiations in the NetNeutrality debate, it answers the question what the appropriate pricing structure is for differentiated services that are proved to be beneficial to the society? Based on the Shapley value solution, the implied compensation structures for potential applications of differentiated services are presented.
These results provide guidelines for ISPs to settle bilateral disputes, for regulatory institutions to design pricing regulations, and for developers to negotiate and provide differentiated services on top of the current Internet.