Despite growing interest in the economic aspects of the Internet, such as network interconnection (peering), pricing, performance, and the profitability of various network types, two historical developments contribute to a persistent disconnect between economic models and actual operational practices on the Internet. First, the Internet became too complex -- in traffic dynamics, topology, and economics -- for currently available analytical tools to allow realistic modeling. Second, the data needed to parameterize more realistic models is simply not available. The problem is fundamental, and familiar: simple models are not valid, and complex models cannot be validated.
This project aims to achieve transformative progress in studying economic aspects of the Internet -- network interconnection (peering), pricing, and the profitability of various network types -- by creating more powerful, empirically parameterized computational tools, and enabling broader validation than previously possible. This project will involve measurement of key properties that impact Internet infrastructure economics, such as interdomain traffic, topology dynamics, routing policies and peering practices. These measurements will serve as inputs to a computational model of network interconnection and dynamics. The investigators will validate the model's ability to reproduce known macroscopic properties of the Internet topology, and its ability to reproduce known historical trends in the evolution of the Internet. The investigators will then use the model to study various "what-if" scenarios relating to interdomain interconnection practices, the stability and dynamics of interdomain peering links, and economic properties of provisioning Internet infrastructure.
The intellectual merit of this project lies in an approach grounded in empirical measurements of macroscopic Internet topology, traffic demand, routing policies, and peering policies. The data promise to reveal important, and thus far elusive, insights into the economic implications of topology dynamics, interdomain traffic characteristics, and routing policy, but they will also inform the parameterization of a model of network interconnection incentives and dynamics.
The broader impact of this project lies in deeper, empirically grounded interpretation of available data on the most opaque sub-discipline of network research -- internetwork economics. The educational side of the project will integrate Internet economics in two Georgia Tech courses, while a PostDoc and a PhD student will graduate as experts in this nascent sub-discipline of Internet research. The data and methods developed during the course of this project will be publicly available and regularly presented to both the research community as well as operator and policy forums, e.g., NANOG, FCC.
Despite much recent interest in economic aspects of the Internet, such as network interconnection (peering), pricing, performance, and the profitability of various network types, two historical developments contribute to a persistent disconnect between economic models and actual operational practices on the Internet. First, the Internet became too complex -- in traffic dynamics, topology, and economics -- for currently available analytical tools to allow realistic modeling. Second, the data needed to parameterize more realistic models -- interdomain traffic characteristics, routing and peering policies and pricing/cost structures -- has simply not been available. The problem is fundamental, and familiar: simple models have limited validity, and complex models cannot be validated. In this research project we made significant progress towards two goals: creating more powerful, empirically parameterized computational tools, and enabling broader validation than previously possible. This research comprised a combination of measurement and modeling techniques. We conducted measurement studies of the evolution of the Internet topology at the AS-level, revealing important overall trends in the Internet AS topology (growth, rewiring, and densification), geographical differences, and topology dynamics associated with different types of players (content, transit and enterprise networks). We used flow-level traffic measurements collected at multiple vantage points to infer statistical properties of the interdomain traffic matrix. We used the measured characteristics of interdomain traffic to create ITMgen, a tool for generating synthetic interdomain traffic matrices that match the measured statistical properties. Finally, we characterized the peering practices of Internet transit providers using publicly available data from PeeringDB. This collected data revealed important, and thus far elusive, insights into the economic implications of topology dynamics, interdomain traffic characteristics, and peering policies. This data has also informed the development, parameterization, and validation of ITER and GENESIS, our models of AS interconnection and dynamics. These agent-based network formation models account for the roles of economics, interdomain traffic flow, and provider/peer selection strategies in shaping the structure of the interdomain topology. We validated these models using publicly available data, and investigated a wide range of "what-if" scenarios related to the possible evolution of the Internet ecosystem. The ITER model shed light on the"flattening" trend in the Internet's interdomain topology, and identified the key factors responsible for that transition. With the GENESIS model, we were able to simulate the network formation process, compute distinct equilibria and to also examine the behavior of sample paths that do not converge, analysis which is not possible with analytical game-theoretic models of network formation. We also used the GENESIS model with game-theoretic analysis to explain the gravitation towards Open peering by Internet transit providers and determine the economic impact of this trend. While we have made substantial progress towards measurement and modeling of interdomain economics -- one of the most opaque aspects of the Internet industry -- our work has also revealed areas where more needs to be done. First, computational models that capture the diversity of the real world -- in this case the Internet infrastructure ecosystem -- are highly complex, and thus cannot reasonably be run at Internet-scale. Scaling up our models by exploiting parallelism and distributed computation is a focus of our future work. Second, evolutionary trends in the Internet ecosystem are both influenced by and influence interdomain economics. For example, recent measurements have illuminated an expanding role of major content providers and CDNs in delivering content. Simultaneously, we see vertical integration of transit, access, and content providers, muddling traditional classifications of Internet entities and raising important questions of whether some of these players could exercise market power. Fifteen years ago the AS topology could more realistically modeled using simple customer-provider or settlement-free peering relationships between network service providers; today's relationships often fall somewhere in between these two relatively straightforward interconnection agreements between providers. Little is known about the prevalence of these more complex interdomain relationships, much less how to incorporate them into interconnection economic models. Our future work will consist of extending the scope of our computational models to account for these, and other factors that influence interconnection economics. We disseminated results from this project widely in publications at scientific journals and conferences -- 4 journal publications and 10 conference publications resulted from this work. These publications have been cited a total of 132 times as of October 2013. We published other interesting tidbits from this research on the CAIDA blog. We began a series of workshops at CAIDA focused on Internet Economics, which included participants from experts in academia, public policy, and industry. We published reports on these workshops (2009, 2011 and 2012), in the ACM SIGCOMM Computer Communications Review (CCR). We have made all our collected measurement data, as well as the source code for our computational models and simulators, available publicly to researchers. Participants in this project made 19 presentations at research conferences, operational venues (NANOG), and policy bodies (FCC).
