Content Delivery over Heterogeneous Networks: Fundamental Limits and Distributed Algorithms

There has been a paradigm shift in the design of communication networks due to the content revolution: the focus has shifted from making a connection to delivering content. Content Distribution Networks carry much of the data of the modern Internet, with video alone occupying nearly 80%, consuming a massive amount of resources. The energy cost of data centers rivals that of medium sized countries. Traditional content distribution network designs have involved centralized capital-intensive infrastructure. For scalable performance, next generation content distribution networks will have to harness the "edge-cloud," consisting of vast numbers of highly capable end user devices such as smart phones, tablets, and e-readers. In this project we address the issues of models, fundamental limits, and storage and delivery algorithms associated with the distributed nature of next generation content distribution networks. The main goal of this project is to develop a comprehensive theory for characterizing the fundamental capacity and performance limits of such networks, and to design the associated content storage and distribution algorithms to achieve these limits.

Content distribution networks generate the vast majority of all internet traffic, and we expect our results to guide the next generation of such networks. In particular, we explore ways to support robust and efficient distributed storage and delivery of content over heterogeneous networks. An integral and key aspect of our proposal is to closely tie the theoretical and conceptual aspects of the project with the practical and real-world engineering issues. This approach is highlighted by building testbeds at each of the universities of the investigators as well as specific to each of the main topics (storage, centralized delivery, and peer-to-peer delivery) of the proposal.

As multimedia content consumption becomes the centerpiece of today's on-line experience, the results of this work are expected to contribute to a seamless end-user experience, while freeing up significant backhaul resources and avoiding the saturation that would otherwise plague today's network infrastructure.