Can Bilateral ISP Peering Lead to Network-wide Cooperative Settlement Yang Cheung, Dah-ming Chiu, Jianwei Huang Department of Information Engineering, The Chinese University of Hong Kong Email: {ycheung6, dmchiu, jwhuang}@ie.cuhk.edu.hk Abstract— The Internet includes thousands of Internet service providers (ISPs) which are interconnected to provide connectivity and service for end-users. Traditionally, the settlement between the ISPs are determined based on bilateral agreements that result from pair-wise negotiations. Although this settlement mechanism is intuitive and easy to implement, it does not encourage network- wide cooperation, as the bilateral charges typically do not lead to a fair division of revenue among all ISPs that are involved in carrying the same flows of traffic. This problem is getting more severe with various emerging new Internet business models. In this paper, we try to determine the existence and realizabil- ity of bilateral prices that can achieve fair revenue division among ISPs. In particular, we use Shapley value as the basis for deriving fair prices. Under a quite general topology and traffic model, we find that there exists prices that make the revenue division under bilateral settlement equal to that calculated under Shapley value. The corresponding “fair price” exhibits several nice and desirable characteristics. Moreover, it could be realized approximately. I. I NTRODUCTION The Internet is the result of interconnecting many networks, each operated by a separate ISP. The Internet service, as enjoyed by the customer of all ISP networks, is however a service provided collectively by all ISPs. A customer of the Internet service subscribes to an access ISP. For this service, the access ISP collects a charge from each subscriber. Since the service is accomplished by multiple ISPs, a fundamental issue is how ISPs should divide up this charge. Based on the design of the Internet, the unit of network service is a packet transported. This is a rather miniscule unit for measuring service, not to mention the task of dividing the charge among multiple parties contributing to the service. By convention (established historically), ISPs charge users monthly on a flat rate basis (like an all-you-can-eat buffet), and settle account monthly on a bilateral basis between ISPs who are connected with each other. There are two most common types of bilateral peering relationships between ISPs. In the first type of peering relationship, one ISP is considered as a transit provider for the other ISP, and the transit provider charges its customer ISP for amount of traffic transit through the provider network. The second type of peering relationship is a totally collaborative one. Two ISPs exchange traffic and deem to benefit from it mutually and forego any charges to each other. Although this all seems a rather sloppy business practice, it keeps the effort in book-keeping to a minimum. This minimalist approach is also totally decentralized, without the need for all kinds of coordination between ISPs. In recent years, new business models (using Internet ser- vices) emerge; for example, Internet content providers (ICPs) are able to generate significantly higher levels of revenues [1]. This prompted re-examination of how the benefits of network services should be divided between various players. One study [2] specifically re-examined the issue of how ISPs should share the total revenues of the Internet service and proposed Shapley values as a potential solution. Shapley value refers to an axiomatically derived formula for fairly dividing a prize among a set of contributors in a general economic setting. By forming an coalition, it is argued in [2] that the ISPs will find more optimal routes and maximize the overall value of the Internet. In this paper, we ask a different question: Can the current bilateral peering between ISPs, under proper pricing strategies, lead to a cooperative settlement as if the ISPs are all in a coalition? This proposition is not entirely unreasonable, since ISPs do realize the positive network externality in the interconnected network. There are actually two parts to this question: a) does there exist fair pricing (defined as prices in bilateral peering that produce Shapley value as settlements)? and b) how feasible and likely ISPs will choose such fair prices in their peering agreements. For a general set of ISP network topologies and traffic models, we show the answer to (a) is true. We then show some preliminary results to (b). The paper is organized as follows. We formulate the system model and settlement model in section II and III. In section IV, we show that in a symmetric network setting, we could obtain the fair prices that produce Shapley value and show some desirable properties of the fair prices in Section V. We further consider how fair prices can be calculated in the more general asymmetric topology in Section VI. In Section VII, we show how the fair prices can be locally approximated without knowing the global network information. We conclude this work in section VIII. II. SYSTEM MODELS A. Topology Model We use an AS-level hierarchical model to capture the essence of the current tier-based Internet. The hierarchical model has been extensively used in related literatures (eg. [3], [4]). We assume that each ISP contains only one AS. 1 Figure 1 For an ISP that contains more than one AS, it is enough to think it as one AS in our model.