The paper introduces Receiver-driven Layered Multicast (RLM), a novel approach to multipoint communication in the Internet, addressing the challenges posed by network heterogeneity and scale. RLM combines layered compression with a layered transmission scheme, allowing receivers to adaptively adjust the number of layers they receive based on available network capacity. This approach manages heterogeneity by locally degrading the quality of the transmitted signal at choke points in the network. The key mechanism of RLM involves receivers joining and leaving multicast groups to determine the optimal level of subscription. Receivers perform join-experiments to test the network's capacity, dropping layers if congestion occurs. The protocol uses a learning algorithm to minimize the frequency and duration of join-experiments, reducing their impact on signal quality. Shared learning among receivers further enhances scalability by allowing them to learn from each other's join-experiments. The paper presents a detailed description of the RLM protocol, including its state machine and state maintenance mechanisms. It also discusses the scalability of RLM through simulations, demonstrating its performance in various network topologies and configurations. The simulations show that RLM can achieve low loss rates and fast convergence times, making it suitable for real-time multimedia applications. Finally, the paper explores the implications of RLM on other components of a multimedia communication system, such as receiver consensus, group maintenance, and fairness. It also discusses the integration of RLM with a layered source coder and its potential for implementation in existing video conferencing tools.The paper introduces Receiver-driven Layered Multicast (RLM), a novel approach to multipoint communication in the Internet, addressing the challenges posed by network heterogeneity and scale. RLM combines layered compression with a layered transmission scheme, allowing receivers to adaptively adjust the number of layers they receive based on available network capacity. This approach manages heterogeneity by locally degrading the quality of the transmitted signal at choke points in the network. The key mechanism of RLM involves receivers joining and leaving multicast groups to determine the optimal level of subscription. Receivers perform join-experiments to test the network's capacity, dropping layers if congestion occurs. The protocol uses a learning algorithm to minimize the frequency and duration of join-experiments, reducing their impact on signal quality. Shared learning among receivers further enhances scalability by allowing them to learn from each other's join-experiments. The paper presents a detailed description of the RLM protocol, including its state machine and state maintenance mechanisms. It also discusses the scalability of RLM through simulations, demonstrating its performance in various network topologies and configurations. The simulations show that RLM can achieve low loss rates and fast convergence times, making it suitable for real-time multimedia applications. Finally, the paper explores the implications of RLM on other components of a multimedia communication system, such as receiver consensus, group maintenance, and fairness. It also discusses the integration of RLM with a layered source coder and its potential for implementation in existing video conferencing tools.