Server scalability can be greatly enhanced with hybrid data dissemination that combines multicast and unicast techniques. Multicast is ideal for popular data. It aggregates multiple unicast data transfers into a single transmission and is thus more scalable than unicast. On the other hand, multicast is inappropriate for unpopular data since it would force unwanted contents on most clients. As a result, hybrid data dissemination would ideally achieve scalable utilization of server and network resources while avoiding the reception of unneeded contents.
In this work, we first present a deployment framework for hybrid data delivery in the Internet. Then, we study a fundamental problem in hybrid data dissemination: document selection, which determines the transfer method that is most appropriate for each data item. We individuate special challenges, such as scalable and robust popularity estimation, appropriate classification of documents, and unpopular large documents. In addition, we examine other critical issues including multicast performance evaluation and application level multicast protocols. Our work contributes to building practical hybrid data dissemination network services.
In this work, we first present a deployment framework for hybrid data delivery in the Internet. Then, we study a fundamental problem in hybrid data dissemination: document selection, which determines the transfer method that is most appropriate for each data item. We individuate special challenges, such as scalable and robust popularity estimation, appropriate classification of documents, and unpopular large documents. In addition, we examine other critical issues including multicast performance evaluation and application level multicast protocols. Our work contributes to building practical hybrid data dissemination network services.