On the Cost-QoE Trade-off for Cloud Media Streaming under Amazon EC2 Pricing Models

Exponential growth of video traffic challenges the current paradigm to stream large amounts of video contents to end users. Cloud computing with elastic resource allocation supported enables cost-effective video streaming with desired QoE requirements. We abstract a new theoretical model from real systems for elastic media streaming by introducing a virtual content service provider that rents cloud resources from cloud service providers with multiple purchase pricing options and provides streaming service to end users. The major objective of the virtual content service provider is to reduce monetary cost with good user QoE. The responsibilities of this role are to resolve two problems under real cloud environments: resource provisioning and acquisition. In this new model, more comprehensive formulation of virtual machine provisioning problem is considered by incorporating with dynamic resource provisioning and resource acquiring with real monetary cost under multiple pricing models. We formulate the virtual machine provisioning problem into a constrained stochastic optimization problem, and design an online algorithm which can obtain approximately optimal solution with explicitly provable performance upper bounds through applying Lyapunov optimization framework. Results from extensive trace-driven simulation experiments demonstrate that our proposed OPT-ORS algorithm can well balance the monetary cost and user QoE. We also find that reserved VM instances are utilized fully to satisfy baseline user demand, ondemand instances are always rent only when flash-crowd user demand happens, and spot instances are rent more frequently than on-demand instances to serve user demand over the baseline due to low prices.