Collaborative Research: Redesigning Video Streaming Infrastructure to Systematically Reduce Mobile Device Video Display Energy Consumption in the Video-rich Era

摘要：本研究的目的是通过重新发明整个视频流基础设施设计，开发新颖的解决方案，系统地降低未来视频丰富的移动设备的能耗。最小化显示能耗具有重要的实际意义，因为它占移动设备能耗的很大比例。这项研究的意义在于，从整个流媒体基础设施的角度来看，这是一种新的整体方法，可以利用显示能耗降低。这种方法是一种范式的转变，从现有的方案中探索每个移动设备上的节能，并将为视频流到移动设备开辟新的研究途径。这个高风险、高回报的项目具有更广泛和变革性的影响，超出了本提案中提出的研究。首先，这项研究与以云为中心的信息技术基础设施的大趋势是一致的，并将为构建未来视频流系统建立一个新的工业服务方向。其次，开发的算法、源代码和测量结果将在网上提供给公众和广泛使用，使工业和学术研究人员都受益。最后，对本科生和研究生的研究训练应培养具有信息技术和计算机行业多学科经验的研究人员和工程师。本研究旨在克服当前单设备处理实践的两个根本缺陷：(1)从全局角度来看，由于接收相同视频流的数千或数百万台移动设备需要重复进行相同的计算，因此节能效率非常低。(2)设备层面的非最优显示能耗降低，因为节省的显示能耗与额外处理所需的能耗之间的内在权衡阻碍了采用复杂的算法。本研究的智力价值在于开发了一种范式转换设计，将每个设备的本地处理转移到基于数据中心的全局处理，同时对视频内容进行智能编码，以实现显示节能流。特别是本项目开展了两个主要的研究方向。为了将显示感知集成到视频流基础设施中，首先开发了数据中心的面向显示的视频处理，以利用视频源特性和不同类别的移动设备，以便生成一系列比特流，每个比特流都具有用于在移动设备上降低显示能耗的嵌入式参数。然后，设计了移动接入网关的上下文感知视频适应，对视频比特流进行转码，以匹配设备观看环境和无线信道状态，从而提高体验质量（QoE）。为了克服视频流数量不断增加导致的数据存储开销和视频缓存效率下降的问题，研究了一种基于语法感知的数据中心视频重复数据删除技术，以降低数据存储开销。在此基础上，提出了一种数据中心辅助的网关通用转码方案，以减少接入网关的视频缓存丢失损失。最后，设计了一种针对视频缓存优化的固态数据存储方案，以充分利用闪存用于视频缓存的潜力。

Abstract: The objective of this research is to develop novel solutions that can systematically reduce the energy consumption of future video-rich mobile devices through re-inventing the entire video streaming infrastructure design. Minimizing display energy consumption is of great practical importance since it accounts for a very high percentage of mobile device energy consumption. The significance of this research is that this is a new holistic approach in which display energy reduction is exploited from the entire streaming infrastructure point of view. Such an approach is a paradigm shift from existing schemes that explore the energy reduction on each mobile device and shall open up new research avenues for video streaming to mobile devices. This high-risk, high-payoff project has broader and transformational impact beyond the research proposed in this proposal. First, this research is well aligned with the grand trend towards cloud-centric information technology infrastructure and will establish a new industrial service direction on architecting future video streaming systems. Second, the developed algorithms, source code, and measurement results will be made available online for public and wide-range usage, benefiting both industrial and academic researchers. Finally, the research training to both undergraduate and graduate students shall produce researchers and engineers with multidisciplinary experience in information technology and computer industries.This research aims at overcoming two fundamental drawbacks of current practice in per-device processing: (1) highly inefficient energy saving from global perspective since the same computation shall be repeated among thousands or millions of mobile devices that receive the same video stream. (2) Non-optimal display energy reduction at device level because the inherent trade-off between saved display energy and the energy needed for additional processing is preventing from adopting sophisticated algorithms. The intellectual merit of this research lies in the development of a paradigm shifting design to move the per-device local processing to data center-based global processing along with the intelligent encoding of the video content for display energy efficient streaming. In particular, two major research directions are carried out in this project. To integrate display awareness into the video streaming infrastructure, display-oriented video processing at data centers is developed first to exploit video source characteristics and different classes of mobile devices in order generate a family of bitstreams, each with embedded parameters for display energy reductions at mobile devices. Then, a context-aware video adaptation at mobile access gateways is designed to transcode the video bitstreams in order to match device viewing environments and wireless channel status for enhanced quality of experience (QoE). To overcome increasing number of video streams leading to significant data storage overhead and degradation of video caching efficiency, a syntax-aware video data deduplication at data centers is investigated to reduce the data storage overhead. Then, a data center assisted universal transcoding at gateways is developed to reduce video cache miss penalty at access gateways. Finally, a solid-state data storage scheme optimized for video caching is designed to fully exploit the potential of using flash memory for video caching.