【作者】 符青云；

【导师】 刘心松；

【作者基本信息】 电子科技大学 ， 计算机系统结构， 2009， 博士

【摘要】 对流媒体服务平台性能影响最大的因素包括：流媒体服务器的最大吞吐量、流媒体服务器的Internet接入速率、流媒体存储系统的最大吞吐量。内容分发网络与缓存服务器将流媒体服务的访问点从流媒体服务器变为靠近用户所在区域的缓存服务器，降低了Internet主干网络带宽波动对流媒体服务质量的影响，同时也降低了对流媒体服务器接入速率的要求。客户端P2P数据共享机制减小了流媒体服务器的访问负载，降低了对流媒体服务器吞吐量与Internet接入速率的要求。而流媒体存储系统作为流媒体服务平台的核心部件却研究较少，流媒体的软实时特性使得流媒体存储系统和一般的高性能存储系统的性能指标存在很大不同，在流媒体服务中直接应用一般的高性能存储系统并不能获得最好的服务性能，必须针对流媒体应用的特性设计对应的高性能存储系统。本文研究面向Internet的大规模流媒体服务所需的存储系统结构，其目标是以较低的成本提供高性能、支持QoS确保机制的流媒体存储服务。主要内容如下：1．分析主要的流媒体服务平台体系结构以及已有的各种高性能存储系统直接应用于流媒体服务时存在的不足，指出了研究面向流媒体服务的高性能存储系统的重要性。2．通过系统仿真手段研究了流媒体服务的负载特性，以此为基础研究了存储系统服务性能与缓存容量、数据访问粒度、磁盘带宽之间的关系，获得了存储节点的设计规则。3．提出了数据条带化分片存储与并行访问相结合的高性能存储系统，通过采用并行元数据服务器结构，获得了比类似系统更高的可用性和更高的性能。4．提出了基于三级目录结构的元数据检索机制，提高了检索性能。通过元数据服务器中的快表机制，提高了热点元数据的查询性能。通过在元数据服务中引入元数据预取机制，减小了影片服务过程中查询元数据信息的次数。通过在元数据服务的使用者中引入元数据信息缓存机制，避免了对同一个数据对象多次请求元数据信息而伴随的开销。5．为了进一步提升存储系统的服务性能，在流媒体服务器与存储系统之间引入了独立的分布式缓存层。通过统一管理分布式缓存层的各个节点的缓存空间，构建了全局可见的数据缓冲池。分布式缓存层显著改善了存储系统的访问性能，特别是热点数据的访问性能。6．提出了一种结合了可变比例区分服务模型、接纳控制、多描述编码的服务质量确保机制，为流媒体存储系统提供了服务质量确保能力。7．针对流媒体服务的运营商向Internet上不同区域的多个存储系统部署影片的需求，提出了一种基于最优化服务带宽分配策略，可以适应Internet网络带宽波动的目标节点协同数据分发机制，通过目标节点相互提供对方所需的数据，提高了影片分发性能，降低了数据源节点的负载。更多还原

【Abstract】 The most important factors that influence the performance of the streaming server platform include: streaming server’s maximum throughput, streaming server’s Internet connect speed, and streaming storage server’s maximum throughput. Content Delivery Network and cache server transfer streaming service’s access point from the streaming server to the cache server which is deployed in the area near the client. This mechanism can reduce the negative effects on the quality of streaming service, which is caused by the fluctuating of the bandwidth of the network, and also can reduce the requirements to streaming server’s Internet connect speed. Streaming server’s load is reduced by P2P data sharing mechanism introduced among streaming clients. Thus the requirements of streaming server’s maximum throughput and Internet connect speed are reduced too. Although the streaming storage system is one of the most important components in a streaming server platform, there are relative few researches on it. The soft real-time characteristic of the streaming applications makes a significant difference between the performance indices of the ordinary high performance storage system and those of the streaming storage system. In general, the ordinary high performance storage system cannot get best performance if it is used directly in the streaming service. So it is necessary to design a special high performance storage system according to the characteristics of the streaming applications.In this thesis, the architecture of the storage system for Internet-based large scale streaming service is studied. The goal is to provide the streaming storage service with high performance, the support of the QoS ensuring mechanism, and affordable system investments. The main results are summarized as follows:1. The shortcomings, which occur when popular streaming system architectures and high performance storage systems are used directly in the streaming service, are analyzed in detail. Based on this analysis, the importance of designing a new high performance storage system for streaming service is pointed out and analyzed in detail.2. The load characteristics of the concurrent streaming traffic are analyzed through system simulations. Based on this research, the relationship between the performance of the storage system and several factors, including cache size, data size and disk bandwidth, is studied, and then the design rules for the storage nodes are obtained.3. A high performance storage system based on the stripped data storage and the concurrent data access mechanism is proposed. By introducing the parallel service ability in metadata server, the performance of this system is superior to other existing systems.4. A three-level directory search method is proposed to improve the index performance. By introducing the fast table mechanism in metadata server, this method improves the index performance of hot metadata query service. Also, this method introduces the metadata prefetch mechanism in metadata query to reduce query frequency in the film service procedure. Besides that, this method uses the metadata cache mechanism to prevent repeatedly querying the metadata information of the same data object in the storage client. Thus the metadata server’s cost is reduced.5. In order to further improve storage system’s performance, the distributed cache layer between the streaming server and the storage system is introduced. By uniformly managing all storage space of the nodes in the distributed cache layer, a global visible cache pool is built. This mechanism improves storage system’s performance greatly, especially for data with hot access frequency.6. A mechanism which integrates variable QoS differentiated services model, access control, and the QoS ensuring mechanism of MDC coding method is proposed. It can provide the ability of ensuring QoS for the streaming storage system.7 According to the requirement of the service provider to distribute video file to different areas in Internet, a data cooperative distribution mechanism based on optimum distribution bandwidth allocation policy is proposed, which is suitable for the fluctuating of Internet bandwidth. By cooperatively providing data to other destination nodes, it can improve video file’s distribution performance and reduce the load of the source node..更多还原