With the prevalence of the cloud computing and storage platforms, many VoD applications and data have been ported to cloud infrastructures completely or partially by video providers. Despite of many remarkable advantages offered by cloud platforms, new techniques are needed to fully exploit potentials promised by these advantages so as to better port VoD applications to cloud platforms.
To address this challenge, a number of recent studies in the literature have proposed various optimizations and scheduling schemes that can be broadly divided into two categories. One category is based on cloud servers and the others is based on video channels since a VoD application usually contains many videos that are referred to as video channels. The former focuses on how to adaptively adjust resource and position of cloud servers while the latter concentrates on adaptively managing the video channels and the streams from them. While the video-channel based techniques can generally manage the use of video channels among DCs in a more flexible manner than managing at the cloud-server level in the cloud platform, what is apparently lacking in the existing video-channel based techniques is a thorough and holistic scheme in their decision-making. Without such a thorough scheme, the optimization techniques often fail to provide the desired data availability and access locality in a cost-effective way.
The research group of Prof. Ke Zhou in Information Storage and Optical Display Division of Wuhan National Laboratory for Optoelectronics devises a mathematical model governing resource scheduling, which describes the relationship among replica placement, QoS constraints, bandwidth assignment and cloud operating costs for VoD applications in the cloud and prove that solving this model is NP-hard. Then, we present a feasible distributed heuristic algorithm, called DREAM(-L), that solves the model and produces a budget solution. And, the time and communication complexity of DREAM are analyzed in theory and experiment.
The experiment results demonstrate that DREAM are able to provide perfect data availability and high access locality, and achieve comparable streaming quality at much lower cloud cost of the existing state-of-the-art algorithms. DREAM and DREAML have lower operating costs, respectively savings 49% and 22% of the cost of the Per-DC algorithm, and 41% and 10 of the cost of the Optimal Load Direction algorithm. The paper about this research work has been published in proceeding of INFOCOM’14.
This research is partially supported by the National Natural Science Foundation of China under Grant No. 61232004, and the National Basic Research Program (973 Program) of China under Grant No. 2011CB302305.