Modeling the internet from an economic perspective

Abstract

Modeling the Internet, especially with the emergence of new usage trends, is a challenging problem for the network research community. Recently, user interactions involving content such as sharing, distributing and retrieving images, music, and videos have come to dominate today's Internet usage. In a sense the Internet usage is content-oriented. Consequently, there have been many ongoing efforts to reconsider and redesign various aspects of the Internet to better support this trend. This calls for more practical and comprehensive approaches to understand the Internet, which is a central problem especially in the areas of protocol evaluation, network design and overall performance. In this dissertation, we focus on network optimization problems related to the Internet centered on such content-dominated usage trend. Using the terminology of today's Internet, we model Internet Service Providers (ISPs), which provide network and content services to their customers, and which operate one or more autonomous systems (ASes) that collectively constitute the global Internet. We identify three key problem domains in this space: from the perspective of underlying infrastructure, how a single ISP forms and optimizes its own network topology; how ASes enter into and optimize peering agreements with one another; and from the perspective of support for content-dominated usage, how multiple ISPs collectively facilitate content distribution by strategically placing content motivated by proposed incentive mechanisms. We argue that, compared to more traditional, purely graph-theoretic approaches, modeling focus on economic motivation better underlies and informs each of these problem domains. This dissertation includes our modeling efforts on the following specific problems: (1) A generative model for ISP topologies that explicitly considers economic and technological constraints; (2) A contract-based model for AS-AS topological peering which captures the different business relationships among ASes; and (3) a model for the cost-benefit analysis of content distribution in a global content-oriented network architecture under various incentive mechanisms. Our proposed methods draw primarily from graph-based combinatorial optimization, but also draw on work in cooperative and non-cooperative game theory. We demonstrate that our methods for the first two problems generate more realistic network topologies than traditional approaches, with more explanatory power. For the third problem, we show by analysis and simulation, that a Shapley value-based incentive mechanism for collaborative content placement becomes increasingly compelling as current trends toward larger and richer content continue.