基于网络测量的大规模混合流媒体视频点播系统关键技术研究

With the development of the Internet, Video-on-Demand(VoD) streaming services have become increasingly prevalent, attracted billions of users from all over the world and contributed to a significant portion of the Internet traffic. Internet VoD systems have evolved from general systems to complex giant systems, which are much larger than traditional information systems. However, due to the scale and complexity, it is difficult to analyze and evaluate VoD systems through traditional research methods. It is urgent to use the Internet measurement to conduct a comprehensive understanding and in-depth analysis of Internet VoD systems. Compared with the traditional CDN or P2P architecture, the hybrid architecture can not only enhance streaming performance and system scalability, but also be cost-effective. In this proposal, we choose PPTV and Xunlei Kankan as our research targets, and plan to conduct a comprehensive measurement study on these two large-scale commercial VoD systems to fully reveal the technical details. Based on the measurement data of content, we plan to content dynamics in peer caches and build a suite of mathematical and data-mining model for characterizing the distribution of content lifetime. We plan to perform a comparative study on these two large-scale systems and summarize design space of hybrid VoD systems. Then we will carry out a comprehensive analysis on the key design issues to describe them effects on the final user experience and server load. Based on the measurement and modeling results, we plan to design new switch algorithm and caching algorithm. Our research results will bring forth important insights for understanding the key technology issues of hybrid VoD systems, building content evolvement model, and optimizing algorithms. Our study will be useful for the development and deployment of future VoD streaming systems.

随着互联网的发展，互联网流媒体视频点播服务不断涌现，占据了互联网的绝大部分流量。由于这类系统的超大规模和复杂性导致传统的研究方法很难对其进行分析和评价，迫切需要使用网络测量对其进行全面理解和深入分析。而混合系统与传统的CDN或P2P方案相比，具有服务质量高，基础设备少和可扩展性强等特点。本项目选取两个超大规模的混合系统作为研究对象，拟对这两个系统进行测量，全面揭示其技术细节。通过对内容分布的测量和挖掘，探寻视频内容的演化过程，建立内容生命周期分布模型。全面比较两个系统测量结果，总结出混合流媒体点播系统的设计空间，分析核心设计要素对系统性能的影响。基于上述测量成果，提出新的切换算法和缓存算法。研究意义在于揭示大规模混合系统的关键技术要素，建立内容演化模型，理论评估和优化系统算法。研究成果能够对流媒体系统的开发和改进提供极有价值的借鉴，并起到举足轻重的指导作用。

混合系统是一种具有广泛应用前景的应用架构。本项目面向流媒体点播这一典型场景，采用网络测量进行研究，研究其核心技术挑战以及探求针对高维数据的通用处理方法。研究内容包括：1）对混合系统进行测量和数据预处理，比较分析基础设施部署，内容分发等机制的异同。通过分析数据特征，提出一种基于混合ABC和CRO的高维数据预处理方法；2）针对数据高维度小样本的特点，研发五种通用特征选择方法，提高了数据分析的准确性；3）分析混合系统的设计要素，评估各设计要素对系统性能的影响。结合混合架构的特点，研发了基于改进和谐算法和地理信息系统的联合优化策略对混合系统的基础设施的部署进行优化。本课题探索混合系统的核心关键技术，解决其中的关键科学问题，为混合系统的推广和应用提供了技术支撑，此外，本项目提出的特征选择数据分析方法具有普适性，为其他相关领域的发展提供了方法支撑。