【作者】 廖荣涛；

【导师】 余胜生；

【作者基本信息】 华中科技大学 ， 计算机系统结构， 2008， 博士

【摘要】 基于对等(Peer-to-Peer,P2P)网络的P2P流媒体技术把组播的功能从网络层移到了应用层,充分利用了网络中各对等节点的网络和计算资源,扩展了系统的服务规模,迅速成为当前流媒体技术的研究热点。其中以数据为中心基于自组织网络的P2P流媒体模型拓扑协议简单,对网络波动具有弹性,是目前最主要的模型。改进和提高该模型的服务质量是研究的重点,为此根据节点缓存数据的特征对其涉及的关键问题作出的改进研究工作如下:在成员节点维护中考虑节点间的缓存数据状况,提出以局部优化为目标的分布式动态节点管理机制——基于同步评估的成员管理(Synchronization Estimationbased Membership Management,SEBMM)。该机制在进行节点选择时考虑网络性能和节点之间的数据同步性,用以降低节点组织中只关心网络性能而导致的数据不匹配问题。以节点间工作集相似性为同步标准的方法出现数据传输停顿的概率高,故对SEBMM模型提出进一步改进:以节点缓冲序列的交集空间比例作为节点的同步性评价标准,在进行对等节点的选择时评估节点间缓存数据的互补性,以提高系统节点之间传输有效数据的总量。实验结果表明,该模型能提高P2P流媒体系统的数据连续性,缩短传输路径,降低延时。节点数据调度传输综合考虑节点即时性能以及数据解码特征,提出自适应动态数据调度(Adaptive Dynamic Data Scheduling,ADDS)算法,以提高数据对象的传输效率和质量。为评估节点的动态传输能力,定义其实时调度性能值为任务负载与数据到达率的函数。并在进行多源选择时,将数据包指派给具有最佳调度性能的节点,提高媒体数据的按时到达率。同时为确保接收端数据回放的质量,提出按数据包的调度价值确定传输序列的机制。该调度价值考虑了数据帧的解码特性以及时间关系。此外,在源节点端引入数据的自动推送策略,与数据驱动型模型固有的拉策略相结合构成混杂数据传输模式,该机制基于实时分配信息推测节点的需求,实现部分数据的主动推送,用以提高网络资源的利用率。对节点暂存数据采用差别缓存,提出基于媒体对象有效性预测的缓存管理(Media Object Validity Prediction based Cache Management,MOVPCM)机制,将缓存在节点空间的流媒体对象的置换几率与其缓存价值相关联,降低对象替换带来的解码损失,提高子节点的缓存命中率。为此以提高有效数据总量为目标建立多源缓存管理数学模型,对该模型去除时间耦合以及进行空间局部化,并通过启发式贪婪算法计算最优置换策略。MOVPCM以预测有效度、解码价值以及缓存代价为系数计算该缓存对象的存储优先级,并以该优先级为基础建立对象的差别缓存机制,具有较大解码价值且请求概率高的数据在缓存中具有相对长的生命周期。分析和模拟的结果验证了MOVPCM算法对提高节点的数据命中率的重要作用。P2P流媒体系统是对其涉及的各个关键模块的整合。由于当前系统普遍采用的紧耦合整合机制使得各个模块之间的关联性大,不利于扩充,也与P2P环境下节点的松耦合关系不匹配。因此,在原型系统的实现中采用基于语义的方法,通过松耦合的构建模式来实施P2P流媒体系统的节点和数据的统一管理,提高了系统的灵活性和可扩展性。更多还原

【Abstract】 The Peer-to-Peer (P2P) network based P2P streaming technique migrates the multicast functionality from network layer to application layer, which makes full use of network and computing resources of all joined peers to extend system scale, has become a hot topic in streaming technique field recently. The data-driven P2P streaming models based on self-orgnazation topology is one kind of main models recently, which have attracted more attention because of its simple topology protocol, and being resilient to network dynamics. Usually the QoS (Quality of Service) improvements play important roles in current research in such P2P streaming systems. So according to the characteristic of buffered data, we focus on some key problems as following:Taking the characteristic of buffered data into account for partnership maintaining, the synchronization estimation based membership management (SEBMM) mechanism is proposed. The SEBMM considered both the network performance and data synchronization among peers on peer selections, well decreases the probability of data mismatch which is introduced by partially considered of the network performance without considering the data synchronization. Further research shows that the peers running SEBMM scheme will fall short of data receiving because of the selection criterion adopted by SEBMM. Hence, we can improve SEBMM in such approaches as: applying the cache series intersection ratio between peers as synchronization indicator, and taking data mutually complementarity into account when selecting active peers, through which it can increase the data transmission gross. The simulation results demonstrate the feasibility and effectiveness of this solution.An adaptive dynamic data scheduling (ADDS) algorithm is introduced, which determines data assignment according to the states of peers in dynamic and data characters for decoding. The ADDS algorithm defines scheduling performance as a function of task overload and data arriving ratio, and assigns the packets to peers with the best scheduling performance. The aim laying behind is to increase the data arriving rate in time. In order to assure the receiver-end of playback quality, all data packets to be scheduled will be sorted in order by scheduling value firstly before assignment and those packets which are in urgency and important for streaming decoding will be given priority to be transmitted. With the inherent PULL strategy, a controlled PUSH strategy based PUSH-ADDS algorithm is adopted at source-end, which will push some packets to receiver forwardly. PUSS-ADDS can make use of spare bandwidth of active peers.The proposed media object validity prediction based cache management (MOVPCM) manages cached media objects differently according to their caching value to decrease the decoding distortion arising from packet replacement and increase cache hit rate. A multi-source cache management model with the aim of increasing the valid data gross is also proposed, and after relaxed and localized, it is adopted to work out the optimized replacement series by greedy method. MOVPCM makes use of the predicted validity, decoding value and caching cost as factors to calculate the priority for object caching, based on which objects with high decoding value and request probability will have longer life time in cache. Analysis and simulation results show the validity of the MOVPCM method.P2P streaming system is a combination of some key modals involved. Because previous system tends to adopt fasten coupled system, and such systems can’t be extended facility, and it mismatches with the loosely coupled P2P environment. So, the prototype system proposed in this thesis adopts a kind of semantic method to achieve peers and streaming data management uniformly through a loosely coupled system construction style, improves the flexibility and expansibility of the system in a certain extent.更多还原