Internet services such as web search (e.g, Google and Bing), social networks (e.g., Facebook), and online video streaming (e.g., Netflix and YouTube) are now an important human activity. Underlying such services is a vast, and increasingly complex system of servers, networks, and software controlled by a variety of entities such as content providers (CPs) who own or generate the content, content distribution networks (CDNs) who distribute that content to users, and Internet Service Providers (ISPs) who own and operate the underlying network infrastructure. These entities form the content distribution ecosystem, each with its own business interest, each attempting to optimize its own cost and performance, more often individually, but sometimes in cooperation with others. This project is focused on understanding the complex interactions among these different entities with the view of providing better architectural solutions to facilitate those interactions. Insights gained from the project can help guide the evolution of future Internet services, resulting in better quality-of-experience (QoE) for the users and greater system efficiencies for the entities in the ecosystem.

This project advances a framework to study complex interactions in the content distribution ecosystem from multiple perspectives -- control information, decision points, time scales and local vs. global decision making, and applies this framework to understand their roles and effects on user QoE. The project is organized along four main threads: 1) Complex interactions between CDNs and ISPs: Characterize effects of CDN actions such as content adaptation and request routing on ISP actions such as traffic engineering and traffic shaping; 2) Complex interactions between CPs and CDNs: Characterize CP actions such as choosing the right CDN to deliver their content with CDN actions such as server selection; 3) Complex interactions between (uni- and multi-path) TCP and applications: Characterize interactions between control loops present in applications such as video streaming and protocols such as (MP)TCP; 4) Measurement, trace-driven evaluation, and controlled and "in the wild" experimentation. All four threads are connected by an analytical thrust focused on the development of theoretical models, advanced algorithms, and performance bounds based on optimization, control and economic methods.

The main premise behind this project is that content distribution is becoming more complex with increasing numbers of entities and varied interactions between them. This complexity requires new analytical and algorithmic methods to model, understand, control, and optimize content distribution. The project takes a first step towards developing these tools and methods, and applying them to interactions that arise between CPs, CDNs, and ISPs, and between different application-level and transport level data transfers. If successful, the project will produce sound theoretical underpinnings and provide general design guidelines in effectively improving the system performance and enhancing overall user QoE and benefit CPs, CDNs and ISPs alike.

The successful completion of this project promises to enhance the economic viability of the content distribution ecosystem and the Internet infrastructure at large. Moreover, it will lead to a better understanding of how to engineer large systems consisting of a multitude of diverse interacting components. Educational activities for this project will enhance undergraduate, graduate, and professional education lying at the intersection of networking, modeling, control theory, and complex systems. The PIs will accomplish these goals through 1) creation of courses and course material on complex systems, content distribution systems, and mathematics of interacting systems appropriate for seniors and graduate students; 2) development of outreach activities aimed at increasing participation of undergraduate women and minority students in research as well as engagement with the broader networking community.

Agency
National Science Foundation (NSF)
Institute
Division of Computer and Network Systems (CNS)
Application #
1413998
Program Officer
Darleen Fisher
Project Start
Project End
Budget Start
2014-10-01
Budget End
2019-09-30
Support Year
Fiscal Year
2014
Total Cost
$1,734,551
Indirect Cost
Name
University of Massachusetts Amherst
Department
Type
DUNS #
City
Hadley
State
MA
Country
United States
Zip Code
01035