TCP-COMPATIBLE RATE CONTROL FOR FGS LAYERED MULTICAST VIDEO TRANSMISSION BASED ON A CLUSTERING ALGORITHM J´ erˆ ome Vi´ eron, Thierry Turletti , Xavier H´ enocq, Christine Guillemot and Kave Salamatian INRIA Campus de Beaulieu, 35042 Rennes Cedex, France firstname.lastname@inria.fr ABSTRACT This paper addresses the problem of rate and congestion control for video transmission in large multicast groups. A classification of the receivers is performed according to a predefined similarity measure (e.g. packet loss rate observed or TCP-friendly bandwidth share). The RTCP receiver reports are then filtered according- ly. The classification of receivers done by feedback aggregation agents helps in solving the feedback implosion problem in large multicast groups. The clustering and feedback aggregation mech- anisms are coupled with an FGS layered video coding system in which the number of layers as well as their respective rates are adapted dynamically to aggregated feedback. The algorithm has been validated via NS 2 simulations. 1. INTRODUCTION In a multicast topology (multicast delivery tree in the 1 case, acyclic graph in the case), network conditions such as loss rate and queueing delays are not homogeneous in the gener- al case. Rather, there may be local congestions affecting down- stream delivery of the video stream in some branches of the topol- ogy. Hence, the different receivers are connected to the source via paths with varying delays, loss and bandwidth characteristics. Due to this potential heterogeneity, dynamic adaptation of multimedia flows over multicast channels, for optimized QoS of multimedia sessions, faces challenging problems. In order to adapt the source and transmission parameters to the network state, one has to rely on the usage of feedback mech- anisms. However, the use of feedback schemes in large multicast trees faces the potential problem of feedback implosion. The first issue addressed here is therefore the problem of aggregating het- erogeneous reports into a consistent view of the communication state. The second issue concerns the design of a source rate con- trol mechanism that would allow a receiver to receive the source signal with quality commensurate with the bandwidth and loss ca- pacity of the path leading to it. A variety of multicast schemes - sender [1] or receiver driven [2] - making use of layered coding have been proposed. Receiver- driven approaches consist in multicasting different layers of video using different multicast addresses and let the receivers decide which multicast group(s) to subscribe to [2]. This approach suffers INRIA, 2004 route des Lucioles - BP 93,06902 Sophia Antipolis Cedex, France. LIP6, University of Paris VI, France. This work has been supported by the french ministry of industry in the context of the national project RNRT-VISI. from some limitations : The first limitation concerns the trade-off between the granularity in the rate adaptation on one hand and traf- fic overhead, extra complexity in multicast address management, and compression penalty on the other hand that would be induced by a large number of layers. A source adaptive multilayered mul- ticast (SAMM) algorithm based on feedback packets containing information on the goodput of the path is described in [3]. Feed- back mergers are assumed to be deployed in the network nodes to avoid feedback implosion. The rate of each layer is chosen in order to maximize a combined ”goodput” measure of video traffic received by all downstream receivers. The goodput is defined as the total throughput of all multicast sessions received without loss, and does not accurately reflect the video quality. In order to solve the feedback implosion problem, [4] proposes a mechanism based on ”partial suppression” of feedback information. This approach avoids the deployment of aggregation mechanisms in the network nodes, but on the other hand, the partial feedback suppression may induce a flat distribution of the requested rates. In order to proceed with the adaptation of the transmission parameters to the network state, the sender does not need reports of each receiver in the multicast group. It rather needs a parti- tion of the receivers into homogeneous classes. Each layer of the source can then be adapted to the characteristics of one class or of a group of classes. Each class represents a group of homoge- neous receivers according to discriminative variables related to the received signal quality. The clustering mechanism used here fol- lows the above principles. A classification of receiver reports is performed by aggregation agents organized into a hierarchy of lo- cal regions. The aggregation agents classify receivers according to similar reception behaviors, and filter correspondingly the RTCP receiver reports. By classifying receivers, this mechanism solves the feedback implosion problem and at the same time provides the sender with a compressed representation of the receivers. This representation of the communication state in the multicast tree is then exploited in order to dynamically adapt the number of lay- ers as well as the rate of each layer of an FGS video source. In contrast with previous work, this adaptation relies on an optimiza- tion performed on the sender side, and which in addition takes into account both the network aggregated state as well as the rate- distortion characteristics of the source. The latter allows to opti- mize the quality perceived by each receiver in the multicast tree. 2. HYBRID SENDER/RECEIVER CONTROL BASED ON AN AGGREGATED FEEDBACK ALGORITHM At the beginning of the session, the sender announces the range of rates (i.e. a rate interval ) estimated from the aver-