Performance-constrained energy reduction in data centers for video-sharing services

Energy saving in large-scale video sharing data centers is an important yet daunting challenge due to the conflicting goal of providing real-time guarantees. Simple energy reduction techniques can result in excessive delay and severely affect the quality-of-service. This paper aims to optimize energy consumption while ensuring service delay constraints in data centers that provide large-scale video-sharing services. However, this broader goal requires three challenges that must be holistically addressed rather than in isolation. First, we propose a generic model to accurately characterize the disk behavior in a VSS by taking into account the unique characteristic of parallel video workloads. Second, the paper proposes a prediction-based algorithm that formulates and solves a constrained optimization problem for determining optimal selections of disk power modes in VSSs. Third, two novel caching algorithms are proposed that achieve additional energy saving through optimizing cache utilization. Experiments reveal that the proposed 3-component scheme achieves a significant amount of energy saving under the same delay level as compared to traditional energy management schemes. A new model for disk idle time in video-sharing services (VSSs) is proposed.We optimize the selection of disk power modes with delay constraints.We propose efficient energy-aware caching and prefetching schemes for VSSs.Significant energy saving is achieved for VSSs under very low service delay.