面向动态随机需求的媒体云自动扩容关键方法研究

With the rapid increase of streaming media content on the Internet, the way of streaming media discovery and dissemination changes dramatically, which brings highly dynamic and stochastic resource demands in media cloud. Existed demand mean-based scaling methods neglect the fine-scale temporal variation of demands, thus lead to inefficient resource allocation. To address the problems introduced by dynamic and stochastic demands, by considering typical characteristics of streaming media application, our research aims for key models and scaling methods in media cloud. It includes: dynamic and stochastic demands oriented resource multiplexing in media cloud, which reduces fluctuations of aggregated demands; stochastic demands oriented offline scaling in media cloud, which can quickly calculate initial resource allocation scheme; dynamic and stochastic demands oriented online auto-scaling in media cloud, which is used to ensure quality of service by scaling in real-time. Our research is expected to provide novel methods on resource multiplexing and auto-scaling in media cloud, the achievements will help cope with dynamic and stochastic demands effectively, further optimize cost when satisfying quality of service.

随着互联网上流媒体内容的快速增加，流媒体信息的传播和发现方式发生巨大变化，媒体云中资源需求呈现高度动态随机特征。现有基于需求均值的扩容方法忽略细时间粒度上的需求变化信息，导致资源配置趋于低效。本项目针对动态随机需求引入的新问题，结合流媒体应用典型特征，研究媒体云扩容的关键模型和方法，具体内容包括：面向随机需求的媒体云资源多路复用，降低聚集随机需求的波动；面向随机需求的媒体云离线扩容，快速获取初始资源配置方案；面向动态随机需求的媒体云在线自动扩容，实时调整资源配置以满足服务质量要求。本项目预期在媒体云资源多路复用和自动扩容等方面有关键方法上的创新，项目成果有助于高效应对动态随机的资源需求，满足服务质量要求并进一步优化费用。

随着互联网流媒体流量的快速增加，社交网络等多种应用规模的持续增长，媒体云资源需求呈现高动态随机特点。现有自动扩容方法忽略需求动态随机等细节特征，无法充分满足流媒体服务质量要求。面对高度随机的资源需求，如何考虑异质费用模型并保证跨区域服务质量，在分布式媒体云中合理分配资源并进行资源多路复用，成为分布式云资源调度亟需解决的关键技术问题。本项目紧密结合流媒体等高通量应用中需求高度动态随机的特点，研究结合应用特征的分布式媒体云自动扩容关键模型和方法，重点研究资源量限制下通用云资源调度方法，基于通用价格函数的费用总额限制下异质云资源调度方法，混合云中处理动态内容分发的资源调度方法，用于云资源调度优化的最大流加速方法。最后基于真实和仿真的数据及实验平台，评价和完善相关理论和算法实现。