QQ客服 官方微博 反馈留言

题名	基于对等网络的流媒体直播系统的关键技术研究
作者	鲁逸峰
学位类别	博士
答辩日期	2009-05-26
授予单位	中国科学院声学研究所
授予地点	声学研究所
关键词	对等网络 视频直播 索引服务器 IP组播 分布式哈希表 系统传输延迟 数据调度 服务调度 频道间协作
其他题名	Studies on Key Technologies of Live Streaming Systems over Peer-to-Peer Network
学位专业	信号与信息处理
中文摘要	流媒体直播服务已成为互联网最重要的应用之一，然而，无论是基于客户端/服务器模式的传统流媒体技术，还是以IP组播与内容分发网络为代表的新兴流媒体技术，都已经越来越难以满足飞速增长的用户需求。对等网络凭借其优异的可扩展性以及低成本、易部署的优势，为流媒体直播服务提供了一个新的发展方向。 本文重点研究基于对等网络的流媒体直播系统的三个关键技术：节点组织方法、数据调度策略以及频道间协作策略，主要贡献与创新工作如下： 1. 提出了一种结合索引服务器与DHT技术的I-DHT协议，并通过其构建应用层覆盖网。通过索引服务器的查询功能，实现节点的并发加入与就近路由，从而提高节点的加入速度与路由速度； 2. 提出了一种结合P2P技术与IP组播的节点组织方法。一方面，通过I-DHT协议构建应用层覆盖网；另一方面，利用IP组播组织组播域内的节点并分发媒体数据。通过结合P2P技术与IP组播，不仅能够充分利用节点的网络资源，减轻数据源服务器与骨干网的负载，提高系统的可扩展性，而且能够有效提高视频流的服务质量； 3. 采用基于推模式的数据调度策略建立数据传输模型，通过理论分析得到系统传输延迟的递推关系式，并计算系统传输延迟与数据源带宽消耗的关系。由于同时考虑了系统直径与单跳延迟的影响，故能够获得更为准确的结果。计算结果给出了特定条件下系统传输延迟的下限，对应用层覆盖网及数据调度策略的设计具有重要的指导意义； 4. 提出了推拉结合的启发式数据调度策略与服务调度策略。在数据调度策略中，基于多级反馈队列对合作节点的服务能力进行评估，并按照评估结果调整调度量，同时，根据数据片段的状态选用拉模式或推模式；在服务调度策略中，优先保证请求节点已获得的服务能力，并将剩余的服务能力公平地分配给各个请求节点。这些策略能够在保证视频流畅播放的前提下，充分利用节点的网络资源，有效降低系统传输延迟。仿真结果表明，与基于拉模式的数据调度策略相比，这些策略能够将系统传输延迟降低一个数量级； 5. 提出了一种基于服务能力的合作节点选择策略。在该策略中，通过选择合适的合作节点，动态调整应用层覆盖网的结构，使得服务能力越强的节点距离数据源服务器越近，从而降低系统传输延迟。仿真结果表明，该策略能够将系统传输延迟降低30%； 6. 提出了一种频道间协作策略。在该策略中，节点通过局部判断算法确定频道状态，如果处于冷播状态，则通过gossip方式寻找具有富余服务能力的热播节点；而对于具有富余服务能力的热播节点，则通过gossip方式主动寻找需要帮助的频道。通过频道间协作，能够充分利用节点的网络资源，从而均衡各个频道的服务能力，提高系统的可扩展性。
英文摘要	With the widespread penetration of broadband accesses, streaming services, especially live streaming services, have become one of the most important Internet applications. However, it’s more and more difficult for current technologies to meet users’ requirement. By exploiting resource of client peers, peer-to-peer technology obtains significant advantages on scalability, robustness and network utility, hence providing a new direction for live media streaming. In this thesis, we make studies on key technologies of live streaming systems over P2P network. Researches mainly focus on: peers organization, data scheduling strategy and cross-channel collaboration strategy. Contributions and innovative work of this thesis are described as follows: 1. I-DHT protocol, which integrates the principles of index server and DHT technology, is proposed. By exploiting the query operation of index server, concurrent joining and nearer routing are implemented, so as to improve the performance of peer joining and routing. 2. A peers organizing method, which integrates P2P technology and IP multicast, is proposed. On one hand, the overlay network is constructed using I-DHT protocol; on the other hand, peers in the same multicast domain are organized using IP multicast to distribute media packets. This method can fully exploit resource of peers to alleviate the load on source servers and backbones, hence improves the system scalability. Moreover, the QoS of video stream is also improved. 3. A data transmitting model based on push mechanism is presented. Based on this model, the recursiveness of system delay is investigated through theoretical analysis, and its relationship with server bandwidth cost is computed. Especially, both overlay diameter and one hop delay are taken into consideration, hence can obtain more exact results. The findings are of great importance for the design of P2P overlay and data scheduling strategy. 4. A heuristic pull-push data scheduling strategy and a service scheduling strategy are proposed. On one hand, the data scheduling strategy evaluates service capacity of cooperative peers based on multilevel feedback queues, and make adaptive adjustment. Meantime, pull or push mechanism is activated to request data chunks in term of their status. On the other hand, the service scheduling strategy firstly guarantees the service capacity that requesting peers have gained, and then divides the left service capacity to all requesting peers equally. On the premise of playing fluency, these strategies can fully exploit resource of peers, hence reduces the system delay. As the simulation results indicate, compared with pull-based strategy, the system delay of pull-push strategy is reduced by one order of magnitude; 5. A heuristic peer selection strategy based on service capacity is proposed. By selecting cooperative peers in terms of their service capacity, the overlay topology is adjusted adaptively, in which peers of greater service capacity will locate nearer the source server. With the strategy, better system delay can be achieved. As the simulation results indicate, the system delay is reduced by 30%; 6. A cross-channel collaboration strategy is proposed. Peers use local judgment algorithm to check whether their channels are cold or hot. Cold peers adopt gossip protocol to find peers of surplus service capacity for collaboration. Similarly, hot peers of surplus service capacity adopt gossip protocol to find channels which need help. Moreover, the collaboration strategy will make adaptive adjustment according to the change of channel status. With the strategy, peers across different channels will collaborate with each other to balance the service capacity of each channel. Consequently, resource of peers can be fully exploited to improve the system scalability.
语种	中文
公开日期	2011-05-07
页码	158
内容类型	学位论文
源URL	[http://159.226.59.140/handle/311008/482]
专题	声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文
推荐引用方式 GB/T 7714	鲁逸峰. 基于对等网络的流媒体直播系统的关键技术研究[D]. 声学研究所. 中国科学院声学研究所. 2009.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。