Towards cost-driven control and management for cloud centric media network

Nowadays, the emergence of cloud computing is transforming the way of operating various IT infrastructures, including video distribution architecture, with an objective to improve their efficiency. Specifically, the cloud-based paradigm enables the video service operators to dynamically leverage virtualized resources, such as bandwidth, computation and storage, so that the user requests can be served in an on-demand manner. As a result, it provides an opportunity to reduce the operational cost of delivering video contents, throughout careful scheduling and management on the usage of cloud resources. In this thesis, we investigate the cost-efficient control and management problem in cloud centric media network, a cloud based media content distribution paradigm. First, we consider the cost optimization problem for platform services. Specifically, we focus on three cases with different combinations of those functions. The first case stands for a typical content delivery system, where the operators can only utilize the cloud caching resource, and decide how to allocate them for different contents and locations. For this case, we mathematically establish the relation between the replica number and the mean hop distances in large-scale networks, and theoretically derive the optimal replica number for each content with different popularity. The second case introduces online transcoding function to media cloud nodes, making it possible to jointly operate caching and transcoding functions for adaptive video streaming. For this case, we develop an analytical framework to balance a three-way trade-off between storage, computing and bandwidth resources, and derive a closed-form solution for cloud resource allocation. Finally, the last case jointly optimizes not only in-network caching and transcoding function but also the network routing function for each node, enabled by the emergence of Software Defined Networking (SDN) and Network Function Virtualization (NFV). For this case, we propose a two-step iterative approach to find the optimal solution, and develop an analytical framework to prove its convergence and optimality. Then, we design and implement a multi-screen cloud social TV system as a novel cloud application on top of the cloud centric media network, and try to minimize the operational cost of such service. Specifically, multi-screen cloud social TV uses cloud clone as its key technology, offering a rich set of functions including video transcoding, session synchronizing, and video teleportation. As a result, users can seamlessly transfer ongoing sessions among heterogeneous media outlets instantly. Under this application scenario, we formulate the cost-minimization problem as a Markov Decision Process by adopting a Markov chain to model the user watching behavior across TV and smartphone. The objective is to minimize the monetary cost of operating the video teleportation service, by migrating the cloud clone to the best place, as the user shifts his device. We further propose four algorithms to address this optimization problem. The aforementioned cost optimization policies, including the optimization processes over both platform and application services, can efficiently reduce the operational cost of distributing media contents. The proposed approaches and obtained results may provide guidelines to improve the efficiency of using cloud resources for video distribution services.DOCTOR OF PHILOSOPHY (SCE