IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 15, NO. 8, AUGUST 2005 961 Robust and Efficient Path Diversity in Application-Layer Multicast for Video Streaming Ruixiong Tian, Qian Zhang, Senior Member, IEEE, Zhe Xiang, Yongqiang Xiong, Member, IEEE, Xing Li, and Wenwu Zhu, Senior Member, IEEE Abstract—Application-layer multicast (ALM), as alternative to IP multicast, provides group communication without the need for network infrastructure support. To improve the reliability of ALM service, path diversity has been studied and two schemes to con- struct diverse paths for hosts are proposed. One is the random mul- ticast forest (RMF) and the other is topology-aware hierarchical ar- rangement graph (THAG). RMF makes the paths from the media source to a participating host diverse by selecting parents for each host randomly, while THAG makes the paths node-disjoint by con- structing multiple independent multicast trees, where any interior node in a multicast tree will be leaf node in all the other multicast trees. Topology-awareness is implemented in both schemes to make them efficient for media delivery. We compare the reliability and efficiency of THAG and RMF through extensive simulation. The results show that the reliability of THAG has been improved up to 20% compared with RMF. The efficiency metrics, such as rela- tive delay penalty, link stress, and delay variation among different trees in THAG, are also smaller than or almost the same as that in RMF. The results indicate that THAG is a reliable and efficient ALM scheme for streaming media service. Index Terms—Streaming media, application-layer multicast (ALM), fault-tolerance, topology-awareness. I. INTRODUCTION W ITH the rapid growth of streaming applications, pro- viding streaming broadcast service across large-scale In- ternet attracts lots of interests in recent years. In all those re- search efforts, application-layer multicast (ALM) becomes the most prospective means since it inherits the effectiveness of multicast in group communications and overcomes the deploy- ment difficulties of IP-layer multicast. There has been much work in recent years on the topic of streaming media over ALM in the literature, such as NICE [1], Narada [2], CoopNet [4], HMTP [5], OMNI [6], SpreadIt [7], and SplitStream [8]. In ALM systems, the multicast tree is rooted at the media server and participating hosts join the tree as interior or leaf nodes. Interior nodes are responsible for forwarding media data from its parent node to its children Manuscript received September 30, 2003; revised March 1, 2004. This paper was recommended by Associate Editor T. Chiang. R. Tian was with Microsoft Research Asia, Beijing 100080, China. He is now with IBM China Research Laboratory, Beijing 100084, China (e-mail: rx- tian@ns.6test.edu.cn). Q. Zhang, Z. Xiang, Y. Xiong, and W. Zhu are with the Microsoft Research Asia, Beijing, China (e-mail: qianz@microsoft.com; xiangzhe@hotmail.com; yqx@microsoft.com). X. Li is with the Tsinghua University, Beijing 100085, China. Digital Object Identifier 10.1109/TCSVT.2005.852416 nodes through unicast. Considering the quality-of-service (QoS) requirements for streaming media applications, such as playback continuity, propagation delay constraint, and large bandwidth consumption, the following issues are critical for media streaming service that need to be addressed. First, the behaviors of participating hosts are unpredictable since the hosts have the freedom to join and leave the service at any time. The departure or failure of any interior node in the multicast tree will severely affect the descendent nodes so that the stability of multicast service will be affected greatly by node dynamics. Second, the propagation delay from media source to participating node may be excessive because the media data is forwarded by a number of interior nodes along multi- cast tree. Since end hosts in ALM do not have the routing in- formation available to routers, the multicast trees built in ALM suffers from the increasing of propagation delay and the ineffi- cient usage of bandwidth compared with IP multicast. Thirdly, the hosts and network infrastructure are heterogeneous in large- scale ALM system. The service capability on an interior node is subject to both the available network bandwidth and processing capability. Moreover, different network links exhibit different characteristics, such as bandwidth, delay, and packet loss, which greatly affect the QoS of streaming service perceived by end users. It is known that Internet is highly dynamic, while ALM based on Internet is much more dynamic because of the host unpredictability. It makes seeking for optimized system design for media transmission excessively complex in ALM. In ALM, each participating host can communicate with another host directly or through one or more other relay hosts, which means that each node has potential multiple paths to communicate with media source. These abundant optional paths provide flexibility to explore methods of improving media transmission performance through path diversity. Using a number of paths simultaneously, the application can effectively see the “average” path behavior. The average path behavior is generally better than the behavior of any individual path, such as increasing of throughput, reducing of packet loss, etc. This is referred as path diversity [18]. The basic intuition be- hind path diversity is that although media transmission in each path may not optimized, if several paths are not sharing the same congested or deteriorated intermediate nodes or links, the av- erage performance on several paths is better than that on any individual path. Utilizing path diversity, the probability of an outage caused by path damage, which is normal in ALM due to host dynamics, decreases dramatically, because here an outage 1051-8215/$20.00 © 2005 IEEE