A study of cloud-assisted strategy for large scale video streaming systems

In recent years, the Internet has witnessed a significant increase in the popularity of video streaming systems for Video-on-Demand (VoD) or live streaming services. The large-scale content distribution of these systems has become increasingly prevalent and contributes to a significant portion of Internet traffic. Designing such a large scale, fast growing video streaming platform with high availability and scalability is technically challenging. Traditionally, it requires either a massive and costly computation, storage and network delivery infrastructure, or a peer-to-peer (P2P) strategy, which deploys local resources of participating users. To make it worse, the servers usually have to be over-provisioned for the peak load to serve the heterogeneous and dynamic user demands in a large scale with guaranteed Quality of Service (QoS). The emergence of cloud computing however sheds new lights into this dilemma. A cloud platform offers reliable, elastic and cost-effective resource provisioning, which has been dramatically changing the way of enabling scalable and dynamic network services for large scale video streaming systems. First, cloud computing can provide an elastic service and scale the provisioned server resource online in a fine granularity. Second, the geo-distributed cloud sites can respond to the globalized request demands with the qualified service. Third, the cloud-assisted strategy is highly compatible with current prevalent P2P video streaming systems in a hybrid solution. Therefore, the cloud-assisted strategies for large scale video streaming systems call for the novel solutions to provide the cost-effective services. In this thesis, we tackle the design issues of cloud-assisted strategies. First, we propose a flexible cloud based provisioning strategy to serve highly time-varying demands in the P2P VoD systems. Then, leveraging the geo-distributed cloud services, we build a prototype of crowdsourced live streaming platform and explore the streaming quality under the influence of different cloud site leasing strategies. Synergizing the P2P and cloud resource, we model the combinational cost problem, and present an optimal solution and a distributed solution toward a scalable and cost effective service over the hybrid VoD streaming systems. Our analysis and experimental results demonstrate the superiority of proposed systematic solutions for the large scale video streaming systems.