Content distribution in overlay multicast

Multicast is an important communication paradigm to support many distribution applications. Although multicast is conventionally considered as an IP-layer functionality, recently proposed overlay multicast appears to be a more promising solution, where end hosts organize themselves into a logical overlay network and relay data to each other via unicast services. This approach revolutionizes the way network applications can be built, since each node in overlay network is an intelligent one that can collaborate and contribute various resources (CPU, storage, access bandwidth, etc.). Our argument is validated by our study on supporting multimedia content distribution via overlay-based solution. We identify two key challenges: on-demand user requests, where different users may request to view the same multimedia content at different times, and high throughput requirement, where the multicast solution is demanded to maintain data distribution structure with high throughput to each user. Our contributions are as follows. First, regarding the on-demand challenge, we propose an overlay-based on-demand media distribution solution. Through analytical and experimental analysis, we exhibit the great potential of overlay-based solution at saving server load and network bandwidth consumption compared to the ideal IP-multicast-based solutions. Second, regarding the high throughput challenge, using multi-commodity flow theory, we establish the theoretical foundation for multi-tree overlay multicast. Based on this foundation, we propose a series of algorithms, which can achieve maximum throughput for multiple sessions, while maintaining weighted max-min fairness among them. Finally, combining the techniques developed to individually address the above challenges, we propose an overlay-based dynamic high-bandwidth content distribution solution. We prove the approximation bound of our solution regarding the optimal throughput, and show that this bound is greatly outperformed when experimenting under various node dynamics and network topologies.