APPLICATION LEVEL SELECTIVE DROP FOR LAYERED VIDEO OVER MULTICAST NETWORKS Qiang Liu, Jenq-Neng Hwang Dept. of Electrical Engineering, Box #352500 University of Washington, Seattle, WA 98195 {liuq, hwang}@ee.washington.edu ABSTRACT This paper presents an approach of router management that is easy to deploy and can improve the performance of existing layered video schemes. The router is configured to selectively drop a packet of the same application instance from its queue when the network is congested, which may be caused by either network dynamic or failed join experiment. Compared with uniform drop, this application level selective drop (ALSD) can increase the received video quality and provide a more stable subscription level for the narrow-bandwidth receivers competing the same bottleneck link with high-bandwidth receivers. We evaluate the promising performance of the proposed ALSD algorithm with multiple layered video schemes through network simulations. 1. INTRODUCTION Cumulative, receiver-driven layered multicast video schemes [1,2] have been proposed to address the heterogeneity and scalability problems of the Internet. In such a layered video system, the video source uses a layered scalable compression algorithm with a layered transmission scheme and the receiver tries to adapt to the dynamic network condition by joining/leaving a layer (i.e., joining and leaving a multicast group). RLM [1] is the first receiver-driven cumulative layered multicast protocol. The behavior of RLM is determined by a state machine where transitions among the states are triggered by the expiration of timers (the join timer and the detection timer) or the detection of losses. RLM uses “ shared learning” to scale with the number of receivers. RLC [2] is a TCP-friendly version of RLM. RLC is based on the source-generated periodic bursts for bandwidth inference and on synchronization points to scale with the number of receivers. On the other hand, priority drop mechanism at the router is considered to be an alternative or a complement to provide a graceful degradation in the presence of packet loss. When congestion occurs, routers selectively discard less important information (i.e., low-priority packets) before more important information (i.e., high-priority packets). Bajaj, Breslau, and Shenker in [4] analyze the merits of uniform versus priority drop for transmission of layered video and concluded that the performance benefit of priority drop is smaller than expected (the maximum performance gain is about 36% as shown in [4]). Their paper specifically considers uniform drop and priority drop as alternatives to RLM [1] and compares the performances of these algorithms. The authors in [1] argued against priority drop in two ways. Firstly, priority drop rewards poorly behaved users since the video quality doesn’t decrease when the requested rate exceeds the bottleneck. Secondly, under priority drop policy, the receivers may not take the benefit of "shared learning" and impair the scalability of the algorithm. In this paper, application level selective drop (ALSD) is proposed to be used together with a layered multicast video scheme. ALSD is one kind of priority drop that the priority preference is only considered within a single application instance. More specifically, when a router decides to drop a packet, the router searches in the queue for a lower priority packet that belongs to the same application instance. If such a lower priority packet is found, it is dropped while the original target packet is saved; otherwise the original target packet is dropped. We believe that ALSD is easier to deploy than a universal priority drop and it can be adopted with most existing layered multicast video schemes to improve the received video quality. The remainder of this paper is structured as follows. Section 2 describes the ALSD. Section 3 contends that ALSD can help, rather than make worse, a layered multicast video scheme. Section 4 shows the comparison simulation results of ALSD applied to various layered video schemes and Section 5 gives the conclusion of this paper. 2. APPLICATION LEVEL SELECTIVE DROP (ALSD) It is difficult, if not impossible, to apply a single priority preference structure to all applications in a network. There are many kinds of applications and it is not clear how to assign priority to the packets belonging to different applications. An application will try to always assign higher priority to its own packets to protect them against other applications. This defeats the original intention to apply priority to the system. However, it is feasible to assign priorities within a single application instance. The application is well aware that which packet is more important so it can assign proper priorities to different packets without the concern that the assignment may be adversely affected by other applications. At the same time, the router also needs to distinguish different applications when it starts to drop packets. There are some mechanisms for the router to V - 768 0-7803-7663-3/03/$17.00 ©2003 IEEE ICASSP 2003 ➠ ➡