Designing Networks for Innovation

The last decades have seen tremendous growth and transformation in the Internet's commercial landscape. Underneath this success, however, the underlying network architecture has shown a marked resistance to change; it is now described as stagnant and ossified. Numerous design proposals have been developed by researchers, implemented in code, and written right into the routers and end-systems of the network, only to languish as network operators fail to activate them on a large scale. This has spawned a deep pessimism in the network research community, which has adapted over time by favoring increasingly incremental research.

Resolving this resistance to evolution will require an array of reforms spanning the technical infrastructure, the system of contracts between providers, and the governing legal regime. These high-level dimensions are closely interconnected, and changes along one axis may yield adverse results unless paired with specific changes in other dimensions. A comprehensive, top-down approach can highlight interdependencies among design choices, building a high-level conceptual map of the network design space, and identifying beneficial responses.

An emerging technique in economics describes technological progress with an ideas model of innovation, and may form a sound basis for an investigation of network evolution. This literature features an explicit separation between a process of idea generation, and decisions by firms as to whether to invest in those ideas. This is well-suited to describing the Internet's resistance to evolution, since the creation of new ideas appears as vibrant as ever; it is only in the investment decisions of ISPs that innovation stumbles. Explaining this disconnect requires a new array of modeling techniques, particularly emphasizing network topology as a critical input, along with route selection and contracting.

Developing such models, this dissertation characterizes four separate obstacles to network evolution: poor accountability, market concentration in access networks, traffic discrimination and other anticompetitive behaviors, and the end-to-end nature of many technologies. Attempts by researchers to enhance network competition can be divided into four high-level strategies: increased accountability, user empowerment, contract system revision, and virtualized testbeds. This work serves as an appraisal of these strategies and identifies specific uses--and combinations--that may enhance evolution.