Distortion optimized and energy-efficient dynamic video scheduling in virtualized data centers

In this paper, we address the problem of energy-efficient and distortion optimized scheduling of video data in cloud data centers. A cloud-switch video scheduling system is presented, in which the control center is responsible for making switch actions to select an appropriate cloud data center to serve users, by judging a switch weight, and the cloud scheduler is in charge of scheduling video data from cloud data center with time-varying workloads to users. The scheduling problem is formulated as a Markov decision process (MDP), aiming at maximizing the overall video quality received at users, meanwhile, minimizing the total energy consumptions of cloud data centers. We consider decoding dependencies among different frame types, and correspondingly define the frame transmission priority, so as to make foresighted decisions of scheduling multiple frames at each time slot. In order to optimize the long-term utilities, a cloud-switch dynamic video scheduling algorithm is proposed by using dynamic programming. Finally, we evaluate the proposed scheduling algorithm with extensive simulation experiments, and demonstrate that our algorithm can provide a best trade-off between video quality and energy consumptions in cloud environments.