From Peers to the Cloud: Utilizing Distributed Resources for Content Delivery and User Collaboration

In this thesis, we tackle the problem of content delivery and user collaboration with emerging Internet technologies. Our investigation starts from peer-to-peer (P2P) sharing with social relations to contemporary cloud computing with flexible resource provisioning. We seek to leverage distributed resources for efficient sharing and collaboration, which leads to a hybrid system design that seamlessly bridges users' local resources to public datacenters. We first explore social-network-based optimizations inpeer-to-peer content delivery. We give solid evidences that long-term social relations can be found and applied to enhance the sharing efficiency in peer-to-peer networks, and present practical implementation strategies for the popular BitTorrent system. We then investigate the performance of cloud-based file synchronization applications and identify the bottlenecks in their system design, in particular, the task interferences. We propose an interference-aware provisioning algorithm, which effectively mitigates the problem. We further examine the users' interactions in state-of-the-art cloud-based distributed interactive applications. We find that, despite the benefit in terms of cost savings and better scalability, the cloud-based deployment greatly increases the users' interaction latency. We demonstrate that a smart assignment algorithms for virtual machines can remarkably reduce such latency. Finally, we present a real-world system design that effectively bridges users' local resources to enterprise cloud platforms. Our measurements as well as system analysis indicate that it serves as a complement of great potentials to enterprise cloud services.