节点文献
大规模流媒体组播传输技术研究
Research on Multicast Technology for Large-scale Streaming Media Transmission
【作者】 王慧;
【导师】 龚正虎;
【作者基本信息】 国防科学技术大学 , 计算机科学与技术, 2012, 博士
【摘要】 随着三网融合业务的迅猛发展,以IPTV、视频点播及网络直播等为代表的流媒体应用不断涌现。这些应用都拥有庞大的用户规模,适合采用一对多的传输方式。组播技术是一种高效的点到多点数据分发手段,能够有效节省网络资源,减少互联网带宽压力。然而,在大规模流媒体组播应用中,组播路由的自适应性、规模的可扩展性和流量传输的可控性等问题,对现有的组播传输技术提出了巨大挑战。因此,研究大规模流媒体组播传输技术具有十分重要的意义,成为网络传输领域学术界和工业界共同关注的热点。本文针对大规模流媒体组播传输的几个关键问题,深入分析了组播路由协议和算法、体系结构以及传输控制等方面的发展现状,提出了一种基于标播的流媒体传输体系结构,重点研究了基于邻居梯度的组播路由技术、基于Bloom filter的可扩展组播转发引擎和基于服务等级的边缘到边缘动态组播流量控制技术,并在此基础上设计和实现了标播交换节点原型,并对标播流媒体原型系统进行了部署和测试。主要贡献包括以下几个方面:1、提出了一种基于标播的适合于大规模流媒体传输的体系结构LMTA。LMTA将IP接入网络和骨干承载网络隔离,根据流量的特点,骨干承载网络采用标播网络或者经典IP核心网络。LMTA主要设备包括本地媒体中心、标播交换节点、核心路由器、客户端和内容服务器等。本地媒体中心LMC是IP接入网络和骨干承载网络之间重要的控制和转发入口,综合了标签控制器、组成员管理器和边缘路由器的功能。LMTA利用标播传输协议LTP对流媒体数据进行标签交换转发。LMTA对不同类型的流量采用不同的骨干承载网络,具有较好的自适应性和可扩展性。2、针对标播网络和经典IP核心网络的组播路由自适应性问题,提出了一种策略可配置的组播服务模型PMSM,分为策略管理平面、路由控制平面和数据转发平面三个平面,能够将策略灵活地扩展到组播服务模型中。基于此服务模型,定义了邻居梯度的概念和转发规则,综合考虑共享组播树路径、链路剩余带宽、跳数等因素。在此基础上,提出了静态组成员的邻居梯度组播路由算法GMR-S和动态组成员的邻居梯度组播路由算法GMR-D。在根据邻居梯度建立组播树时,利用发现消息和反馈消息进行路由计算。模拟实验显示了该算法的有效性和灵活性。与其他经典的组播算法相比,GMR-S的呼叫受阻率最低,同时组播树代价和平均跳数具有良好的性能,GMR-D能够有效地适应组播树的改变并优化组播树。3、针对经典IP核心网络中核心路由器组播转发可扩展性问题,提出了一种基于Bloom filter的可扩展组播转发引擎BSM。BSM采用Bloom filter保存组播转发信息,路由器的组播信息建立、删除和更新由组成员管理协议来完成,BSM用(s,G)来标识一个组播组,其中s为源的单播地址,G为由源分配的标准的D类组播地址。组播转发信息保存在接口的Bloom filter中,当组播数据报文到达路由器接口时,对报头中的组标识(s,G)进行hash计算,如果匹配成功,则从相应的接口转发。模拟实验表明,BSM不仅可以支持大规模组数量,同时也支持大规模长时间在线的组成员,具有较高的转发效率和较低的带宽消耗。4、针对组播流量的可控性问题,提出了一种与骨干承载网络无关的基于服务等级的边缘到边缘动态组播流量控制机制E2E-DFCM。发送端根据报文对视频质量的影响在报文标记上质量影响标识,发送端视频网关对组播报文进行分类、服务等级映射和标记,并周期性地向接收端视频网关发送前馈报文,接收端视频网关将延迟和报文丢失率等信息反馈给发送端视频网关。发送端视频网关根据网络状况和接收端服务等级的变化动态地调整发送速率。模拟结果显示E2E-DFCM能够有效地调整发送速率,满足用户服务等级的异构性,适合大规模流媒体组播传输。5、基于上述关键技术的研究,设计并实现了基于NetMagic平台的标播交换节点原型。NetMagic有一个内置的用户模块UM,提供了灵活的硬件逻辑可重构功能,通过对UM模块的设计来实现对LTP报文的处理。标播交换节点在处理LTP报文时主要进行修改报文头部的时间戳域和TTL域、重新计算IP头校验、查找标签表获取下一跳输出并替换报头的标签域等处理。同时,实现并部署了标播流媒体传输原型系统。通过建立真实的实验环境对流量进行监测,根据监测结果选择合适的路径进行转发。原型系统的测试验证了流媒体传输体系结构LMTA的可行性和有效性。标播流媒体传输原型系统已经在部分商用IPTV传输平台部署,完成了总体设计、详细设计以及关键技术的实现。综上所述,本文研究工作针对大规模流媒体组播传输中存在的自适应性、可扩展性和可控性方面存在的问题,围绕流媒体组播传输技术展开研究,对于推进组播在流媒体传输中的实际部署具有一定的理论意义和应用价值。
【Abstract】 With the rapid development of Triple Play industries, more and more streamingmedia applications such as IPTV, Video-On-Demand and webcast, etc., arecontinuously emerging. These applications have immensely large users scale, and arevery suitable for point-to-multipoint transmission mode. Multicast is a technique used tofacilitate these types of one-to-many data delivery, by transmitting the same data fromone source to a potentially large number of destinations. So multicast can efficientlysave network resource and reduce the bandwidth stress of Internet. However, with thecontinued growth of streaming media applications, multicast confronts with adaptability,scalability and controllability challenges. Therefore, research on multicast transmissiontechnology is great significance for large-scale streaming media applications, and it hasbeen widely recognized by both global academia and industry today that how to designefficient multicast transmission technology is one of the hot research topics.In this thesis, we study some key problems of large-scale streaming mediamulticast transmission technologies, and argue the recent proposals of multicast routingprotocols, algorithms and architectures. We start our research from the proposedLabelcast based Media Transport Architecture, and then we focus on the research onneighbor gradient based multicast routing technology, a Bloom filter-based scalablemulticast forwarding engine and a service lever based edge-to-edge dynamic multicastflow control technology. We also design and implement a Labelcast switch nodeprototype, and deploy a Labelcast streaming media prototype system to validate ourwork.The major contributions of our work are as following:1. A Labelcast-based Media Transport Architecture (LMTA), which is verysuitable for large-scale streaming media transmission, is proposed. The IP accessnetwork and backbone network is isolated in LMTA. And the Backbone network isselected as Labelcast network or classical IP core network according to the flowcharacteristics. The main equipments in LMTA include: Local Media Center (LMC),Labelcast Switch Node(LSN), Core Router(CR), Client and Content Server. LMC is theimportant controlling and forwarding point between IP access network and backbonenetwork, and it has the integrated functions of Label Controller, Group Manager andEdge Router. Meanwhile, LMTA makes use of Labelcast Transport Protocol(LTP) todelivery data with label switching. LMTA could only adopt different backbone networkfor diverse flow types, but also has the adaptability and scalability in advance comparedwith other multicast architectures.2. Aiming at the adaptability problem of multicast routing in Labelcast networkand classical IP core network, we present a Policy-enabled Multicast Service Model(PMSM), which is divided into three planes: Policy Manage Plane, RoutingControl Plane and Data Forwarding Plane. In PMSM, policies can be embedded flexiblyinto this multicast service. Based on this model, the neighbor gradient definition andforwarding rule is defined, which is calculated based on the weighted sum of attributessuch as residual link capacity, normalized hop count, etc. Then two distributed multicastrouting algorithms which are neighbor Gradient-based Multicast Routing for Staticmulticast membership (GMR-S) and neighbor Gradient-based Multicast Routing forDynamic multicast membership (GMR-D), are proposed. Discovery message andfeedback message are used for discovering multicast routing path when establishing themulticast tree based on the neighbor gradient. GMR-S is suitable for static membershipsituation, while GMR-D can be used for the dynamic membership network environment.Experimental results demonstrate the effectiveness and efficiency of our proposedmethods.3. Aiming at the scalability problem of multicast routing in classical IP corenetwork, we present a Bloom filter-based Scalable Multicast—BSM. In BSM, themulticast information of the BSM router is setup, deleted and updated by membershipmanagement protocol. A multicast group is identified by a group tag(s,G), where s is thesource address and G is the D class IP multicast address. Bloom filter is used to presentthe multicast information in each interface of the routers. When a multicast packetarrives at the router, the group tag(s,G) in the header will be hashed by the hashfunctions. If the results are matched, the packet will be forwarded toward this interface.Simulation results show that BSM can not only support hundreds of thousands ofmulticast groups with long-lived membership, but can also support large multicastgroup size. Meanwhile, BSM can achieve high forwarding efficiency and lowbandwidth overhead.4. Aiming at the controllability problem of multicast transmission, a service leverbased Edge-to-Edge Dynamic Flow Control Multicast (E2E-DFCM), which isindependent from the backbone network, is proposed. The main idea of E2E-DFCM isas follows: The packets which have been labeled with Quality Effect Identifier (QEI) isgathered and remapped by Sending Video Gateway (SVG). Receiving Video Gateway(RVG) periodically sends multicast flow status information to SVG, including delay,loss rate, etc.; SVG dynamically remaps and classifies packets according to QEI andflow status information feedback from RVG, and reassigns each QoS level flow ratesbased on their utilities. Once the QoS of receivers can not be satisfied (i.e.,average delayand loss rate exceed the tolerable range), RVG even requests to down-grade its level,and SVG will reduce the sending rate. In the end, simulation results show thatE2E-DFCM can effectively adjust the sending rate dynamically and satisfy the diversityof user service lever for large-scale streaming media multicast transmission.5. Based on the above research work, a Labelcast switch node prototype based on NetMagic experimental platform is designed and implemented. There is a UserModule(UM) in NetMagic platform, which can provide hardware logic reconfigurablefunction. And the LTP packet processing is implemented by the design of UM. When aLTP packet is arriving at the Labelcast switch node, the packet processing includes:modifying the time and TTL field, recomputing the IP header checksum, looking up thelabel table, getting the next hop port, and replacing the label field, etc. Meanwhile, aLabelcast steaming media transmission prototype system is built and deployed. Thestreaming media flows are monitored in the real experimental environment, and thesuitable paths will be chosen according to the monitor results. The prototype systemvalidates the feasibility and effectiveness of LMTA. Labelcast steaming mediatransmission prototype system has already deployed in some commercial IPTVtransmission platform, and the implementation of key technologies has been achieved.In summary, we focus on the adaptability, scalability and controllability problemsfor large-scale streaming media transmission technologies. We believe that these workshave academic and practical value for advancing the theory and practicability of theabove research.
【Key words】 Multicast; Large-Scale; Streaming Media; Triple Play; IPTV; Adaptability; Scalability; Controllability; Routing Algorithm; Flow Control; Bloomfilter;