Quality-consciousness in Large-scale Content Distribution in the Internet
MetadataShow full item record
Content distribution is the primary function of the Internet today. Technologies like multicast and peer-to-peer networks hold the potential to serve content to large populations in a scalable manner. While multicast provides an efficient transport mechanism for one-to-many and many-to-many delivery of data in an Internet environment, the peer-to-peer networks allow scalable content location and retrieval among large groups of users in the Internet. Incorporating quality-consciousness in these technologies is necessary to enhance the overall experience of clients. This dissertation focuses on the architectures and mechanisms to enhance multicast and peer-to-peer content distribution through quality-consciousness. In particular, the following aspects of quality-consciousness are addressed: 1) client latency, 2) service differentiation, and 3) content quality. Data analysis shows that the existing multicast scheduling algorithms behave unfairly when the access conditions for the popular files changes. They favor the popular files while penalizing the files whose access conditions have not changed. To maintain the client latency for all files under dynamic access conditions we develop a novel multicast scheduling algorithm that requires no change in server provisioning. Service differentiation is a desirable functionality for both multicast and peer-to-peer networks. For multicast, we design a scalable and low overhead service differentiation architecture. For peer-to-peer networks, we focus on a protocol to provide different levels of service to peers based on their contributions in the system. The ability to associate reliable reputations with peers in a peer-to-peer network is a useful feature of these networks. Reliable reputations can help establish trust in these networks and hence improve content quality. They can also be used as a substrate for a service differentiation scheme for these networks. This dissertation develops two methods of tracking peer reputations with varying degrees of reliability and overheads.