Date of Award

Spring 2012

Project Type


Program or Major

Computer Science

Degree Name

Doctor of Philosophy

First Advisor

Radim Bartos


Over the past few years, a revolution in video delivery technology has taken place as mobile viewers and over-the-top (OTT) distribution paradigms have significantly changed the landscape of video delivery services. For decades, high quality video was only available in the home via linear television or physical media. Though Web-based services brought video to desktop and laptop computers, the dominance of proprietary delivery protocols and codecs inhibited research efforts. The recent emergence of HTTP adaptive streaming protocols has prompted a re-evaluation of legacy video delivery paradigms and introduced new questions as to the scalability and manageability of OTT video delivery.

This dissertation addresses the question of how to enable for content and network service providers the ability to monitor and manage large numbers of HTTP adaptive streaming clients in an OTT environment. Our early work focused on demonstrating the viability of server-side pacing schemes to produce an HTTP-based streaming server. We also investigated the ability of client-side pacing schemes to work with both commodity HTTP servers and our HTTP streaming server. Continuing our client-side pacing research, we developed our own client-side data proxy architecture which was implemented on a variety of mobile devices and operating systems. We used the portable client architecture as a platform for investigating different rate adaptation schemes and algorithms. We then concentrated on evaluating the network impact of multiple adaptive bitrate clients competing for limited network resources, and developing schemes for enforcing fair access to network resources.

The main contribution of this dissertation is the definition of segment-level client and network techniques for enforcing class of service (CoS) differentiation between OTT HTTP adaptive streaming clients. We developed a segment-level network proxy architecture which works transparently with adaptive bitrate clients through the use of segment replacement. We also defined a segment-level rate adaptation algorithm which uses download aborts to enforce CoS differentiation across distributed independent clients. The segment-level abstraction more accurately models application-network interactions and highlights the difference between segment-level and packet-level time scales. Our segment-level CoS enforcement techniques provide a foundation for creating scalable managed OTT video delivery services.