TPRC-07: Complexity of Internet Interconnections: Technology, Incentives and Implications for Policy

End-to-End (E2E) packet delivery in the Internet is achieved through a system of interconnections between many different types of entities called Autonomous Systems (ASs), entities that have ownership and administrative control over their networks. Historically the motivation behind and realizations of these interconnections began simple but have become significantly more abundant and complex than is commonly understood. For instance, large content networks, with mainly outbound traffic, are emerging which changes the nature of interconnections. As a consequence settlements are evolving with different allocative goals and mechanisms. The current models fail to accurately reflect the diversity of these and other possible interconnections among ASs that comprise today's actual Internet. In this paper, we present a more accurate picture of what Internet interconnection looks like today and discuss the implications of this for interconnection incentives, contractual arrangements, and public policy.

Specifically, as of March 2007 IANA had allocated 48,126 AS numbers, representing a significant feasible set of interconnections. However, besides the obvious geographical constraints, further preferential, architectural and settlement constraints   induce a relatively smaller possible set of interconnections than the feasible set.   An AS is most often an ISP but can also be enterprises, governmental or educational institutions and increasingly large content providers with mostly outbound traffic such as Google, Yahoo, YouTube and overlay networks such as Akamai, Savvis and Limelight. Each AS is also autonomous and controls and administers its own domain but must physically interconnect with other networks to provide services. A key problem discussed in this paper is the role of preferences among different types of ASs that induce the system of interconnections, or who has both ex-ante and ex-post incentive to interconnect to who given the different AS types. If interconnection is mutually stable then the fixed costs of a physical interconnection are then committed by investing in the physical links or collocations. Each AS then strategically uses the Internet routing algorithm (the Border Routing Protocol, BGP) to control usage-based costs by managing how packets exit and enter their network.

Not surprisingly, the challenge of recovering the fixed and usage-sensitive costs of E2E transport have given rise to a more complex settlements mechanism than the simple bifurcated model of tran sit (where �gsmall�h networks pay �glarger�h ones) and peering (revenue-neutral interconnection among similar size networks). We will demonstrate the emerging developments in this new settlement mechanism, emphasizing their ( ex-ante and ex-post ) difference from settlement mechanisms employed in the telephone network, in which interconnection is heavily regulated and significant carrier resources are devoted to the metering and billing of intercarrier payments.

In particular, we will use case histories, simple formal models and design principles to demonstrate: 1) how the architectural choices motivated by technical desiderata have induced the currently implemented ex-post settlement mechanism, 2) how the two standard settlement mechanism based on a system of bilateral transfers fails to send relevant price signals that can be used to internalize potential indirect externalities across markets, leading to closure of viable markets, 3) the arbitrarily complex ex-ante and ex-post interconnection strategy space available to ASs of different types within that mechanism and 4) how the evolving interconnection preferences of different types of ASs has led to the emergence of potentially more dominant and non-standard settlement mechanism in the market that further extend the strategy space of ASs.

The outcomes of this work will be relevant to not only Internet architects, but also the policy community (e.g., those engaged in debates over Network Neutrality) by highlighting how Internet architecture constrains the options and complexity of interconnection mechanisms and policy.

NETECON-07: Economics of Overlay Networks: An Industrial Organization Perspective on Network Economics

We use theory of Industrial Organization to demonstrate the demand and cost conditions on entry and scaling incentives of Internet overlay networks, restricting the problem to Internet content distribution. We describe the market structure, the nature of demand and the cost-allocation mechanism in both wholesale and retail markets. We show how the end-to-end coordination failures of the ISPs and content providers has resulted in wholesale market failures, inducing entry by propriety content distribution �gmiddleboxes�h that act as intermediaries, coordinating unmet demands. We also show that the intermediary has incentives to strategically use the Internet cost-allocation mechanism to internalize and scale using these indirect externalities from trade. The scaling incentives of such overlay markets is also briefly compared with bargaining institutions implemented by P2P content networks. [.pdf]

In recent years, we have seen the emergence of numerous types of so-called "overlay" networks in the internet. There are many diverse examples of such overlay networks including the content-delivery-caching networks, implemented by companies like Akamai, the peer-to-peer file sharing networks associated with applications such as BitTorrent, the voice-over-IP services offered via Skype, and various testbed networks such as PlanetLab. Such overlays have important technological and policy implications for the evolution of next generation internet architecture. This paper provides a first attempt to understand the implications of such overlays for internet architecture, industry structure, and policy. We introduce a taxonomy for thinking about these overlays with some examples of their scale and growing importance in the internet, and suggest some preliminary thoughts on the implications of these overlays for industry structure and policy. [.pdf]

HotNets-06: Interconnection Discrimination: A Two-Sided Perspective

We first motivate a methodological approach to network economics research using tools from Industrial Organization (IO). We then demonstrate the utility of such models in light of an increasingly important problem of interconnection failures that includes network neutrality problem. In particular, we focus on a subset of the problem by focusing on direct interconnection failures when ISPs price discriminate between content providers and content users. We informally show how viewing the problem as a Two-Sided Market can lead to a discrimination rule which highlights the structural/distributional nature of the interconnection problem. Structural consideration has not only policy and antitrust consequences but can also be highly informative to engineering choices. [.pdf]

NetEcon-06: Assessing the assumptions underlying mechanism design for the Internet

Mechanism design seeks to create incentives that align the interests of selfish agents with the interests of a principal designer. Critical to aligning these interests is the principal's assumptions about the agents that will be playing the game induced by the mechanism. In this paper, we critically examine two classes of assumptions in mechanism design for the Internet: 1) assumptions about agent preferences and 2) assumptions about the number of times an agent plays the game induced by a mechanism. We appeal to both methodological arguments and general theoretical results in exploring the limits and strength of claims in networking that are based upon mechanism design. We demonstrate how previous research [7,15] proves properties that may not hold true on the actual Internet if selfish agents do not conform to a principal's assumptions.

NIC-06: Annealing Protocol for Negotiating Complex Contracts

Work to date on negotiation protocols has focused almost exclusively on defining contracts consisting of one or a few independent issues and relatively small number of possible contracts. Many real-world contracts, by contrast, are much more complex, consisting of multiple inter-dependent issues and intractably large contract spaces. This paper describes a simulated annealing based approach appropriate for negotiating such complex contracts that achieves near-optimal social welfare for negotiations with binary issue dependencies.

TPRC-05: The Growth of Internet Overlay Networks: Implications for Architecture, Industry Structure and Policy

Over the past several years, we have seen the emergence of numerous types of so- called "overlay" networks in the Internet. There are many diverse examples of such overlay networks including the content-delivery-caching networks, implemented by companies like Akamai, the peer-to-peer file sharing networks associated with applications such as BitTorrent, the voice-over-IP services offered via Skype, and various testbed networks such as PlanetLab. These overlay networks enhance or modify the basic functioning of traffic handling within the Internet. Overlays exist in the blurry boundary between what we think of as "the Internet" (a globally interconnected network of IP networks) and the applications that exist on top of the Internet. Overlays also blur the boundaries between the network edges (what we think of as being associated with customer end-nodes) and the network core (what we think of as associated with the services that support the Internet). As such, overlays have important technological and policy implications for the evolution of next generation Internet architecture that historically has been based on the so-called "end-to-end" principle ([SRC84], [BC01]) which relied on a relatively clear demarcation between applications and network services, and edge and core responsibilities. Because of the Internet's growing role as basic infrastructure and increasingly central role in the communications industry, and hence, obvious focus for regulation, changes in Internet architecture have important policy and industry structure implications. For example, from a regulatory perspective, the debate over overlays in Internet space is analogous to the on-going debate over "layered regulation" (see [Fre02], [Sic02], [Wer02]) and how one might distinguish between "basic telecom services" (which may be regulated under Title II as common carriers) and "enhanced" or "information" services (which are unregulated). From an industry structure perspective, overlays are relevant to the tussle over what services are provided by ISPs as opposed to other third party service providers (so-called ESPs) or by customer-managed layers. Since these overlays are important to the future industry structure of our communications industry and its regulation, including issues of VoIP, intercarrier compensation, universal service, and market-power regulation, we need a clearer understanding of what overlays are and might be. This paper provides a first attempt to help frame the issue. Herein we introduce a taxonomy for thinking about these overlays with some examples of their scale and growing importance in the Internet, and we suggest some preliminary thoughts on the implications of these overlays for industry structure and policy. [.pdf]

PINS-05: Practice and Theory of Incentives in Networked Systems [.pdf]

AAMAS-04: Learning User Preferences for Wireless Services Provisioning