基于邻居推荐策略的CDN-P2P流媒体传输系统设计与实现

发布时间：2018-05-06 06:07

本文选题：CDN + P2P　；参考：《北京邮电大学》2014年硕士论文

【摘要】：近些年随着互联网和流媒体技术的发展,流媒体服务逐渐成为互联网的重要应用,并且趋向于向大规模、高质量的方向发展。大规模流媒体服务也成为近年来的研究热点。相关研究表明,结合CDN和P2P技术的优点,将它们应用到流媒体传输系统中能够显著提高流媒体传输的性能。本文重点研究了以下三方面的内容：第一,针对系统中的网络节点具有高度的自治性、动态性的特点,为了使系统中的服务请求节点能够获得更稳定流畅的流媒体传输服务,我们在系统中加入了基于网络距离和稳定性的邻居推荐功能。它能够根据节点的网络距离划分网络域,并帮助节点从网络域中选择稳定性较好的节点作为邻居节点。可以使稳定性高的节点充分匹配,提高了服务的质量。第二,随着系统中节点数量和资源数量增多,我们在系统中增加了基于用户邻域的推荐功能,帮助系统中的用户找到他们可能喜欢的资源。但仅仅依靠中心服务器的协同过滤,计算量大、计算时间长。因此我们在原有的推荐算法上进行了改进,形成了基于兴趣子网的分布式协同过滤推荐算法。该算法将大量的相似性计算任务分配给节点来完成,节点通过自发的相似性计算,组成兴趣子网。然后由中心服务器完成推荐。降低了推荐过程的计算量,又不失推荐准确性。第三,系统中的节点缓存有限,节点只有在缓存了服务请求节点请求的媒体段时,才能向该服务请求节点提供传输服务。显然,能够提供传输服务的节点越多,服务节点的压力就会越小。针对这一问题,我们在系统中加入了基于资源热度和分段优先级的FIFO缓冲区替换策略,该策略根据节目热度和用户观看节目时的习惯来管理缓冲区,优先保留优先级较高的视频块,提高了能够提供服务的节点的个数,减小了服务节点的压力和启动延时。
[Abstract]:In recent years, with the development of Internet and streaming media technology, streaming media service has gradually become an important application of the Internet, and tends to the direction of large scale and high quality. Large-scale streaming media service has also become a research hotspot in recent years. Related studies show that the application of CDN and P2P technology to streaming media transmission system can significantly improve the performance of streaming media transmission. This paper focuses on the following three aspects: We add neighbor recommendation function based on network distance and stability in the system. It can divide the network domain according to the network distance of the node, and help the node to select the stable node as the neighbor node from the network domain. The nodes with high stability can be fully matched and the quality of service can be improved. Secondly as the number of nodes and resources increases in the system we add the recommendation function based on the user neighborhood to help the users to find the resources they may like. However, it only depends on the collaborative filtering of the central server, which results in a large amount of computation and a long calculation time. Therefore, we improve the original recommendation algorithm and form a distributed collaborative filtering recommendation algorithm based on interest subnet. The algorithm allocates a large number of similarity calculation tasks to nodes, which form subnets of interest through spontaneous similarity calculation. The recommendation is then completed by the central server. The calculation amount of recommendation process is reduced, and the accuracy of recommendation is not lost. Thirdly, the node cache in the system is limited. Only when the media segment requested by the service request node is cached, can the node provide the transmission service to the service requesting node. Obviously, the more nodes that can provide transport services, the less pressure on service nodes. In order to solve this problem, we add a FIFO buffer replacement strategy based on resource heat and segment priority, which manages the buffer according to the program heat and user's habit of watching the program. The priority is to retain the high priority video blocks, which can increase the number of nodes that can provide services and reduce the pressure and startup delay of service nodes.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.02;TP391.3

【参考文献】