State scalability and group modelling of multicast in the internet
Jun‐Hong Cui, Mário Gerla · 2003
Multicast is a mechanism to support multi-point communications. By using a tree structure, IP multicast is resource-efficient in delivering data to a group of members simultaneously. However, despite more than one decade's research and engineering efforts, IP multicast is still far from being widely deployed in the Internet. Among the issues which delay its deployment, state scalability is probably the most critical one. When the number of concurrent active groups is large, conventional multicast protocols face severe state space and control traffic overhead problems: large amount of multicast state must be kept at routers, which translates into large memory requirement and slow packet forwarding; and large numbers of multicast trees must be set up and maintained in the network, which means large control overhead. In QoS multicast provisioning, the problem is even worse, since not only the forwarding state but also the resource requirement of a multicast group must be kept at the router. Another critical issue delaying IP multicast deployment is that multicast research has traditionally been plagued by a lack of realistic underlying data and traffic models and an absence of a systematic simulation methodology. The models are critical for the effectiveness of simulation experiments which in turn guide the design of multicast protocols. In this dissertation research, we first address the issue of multicast state scalability. A novel scheme, called aggregated multicast, is proposed to solve the state scalability problem. The key idea is that multiple groups are forced to share a single delivery tree. Using this scheme, not only can the multicast state be reduced significantly, but also the multicast tree management can be simplified dramatically. Based on aggregated multicast, two protocols have been developed for source specific multicast (SSM) and any source multicast (ASM). To provide scalable QoS multicast support, this research work also proposes an architecture, called AQoSM, which employs the aggregated multicast idea and Diff-Serv capabilities to provide QoS multicasting. The second part of this dissertation research work focuses on multicast group membership modelling. In this part, by examining actual multicast group behavior measured in the Internet, we characterize the key features of multicast group membership. Then we propose a comprehensive framework, called GEM, which helps to model the key features of multicast group membership in a realistic fashion. Using GEM, the group spatial properties can be consistently recreated in simulation experiments. Realistic, consistent membership models are critical for a meaningful evaluation and comparison of different multicast routing, QoS provisioning and reliable transmission protocols. Thus, GEM is provided as a contribution to the community with the hope that it can be part of a realistic and systematic multicast evaluation methodology.