Research On Multicast Technologies In Large Scale Group Communications

With the enormous advance in network computing and communication technologies, a diverse range of applications that inherently require group communication and collaboration have arisen. These applications include video/audio conferencing, computer collaboration, larger-scale distributed simulation, etc. The diversity of these applications demands versatile support from the underlying network in many dimensions. In order to support these applications in our current IP networks, researchers have been resolved in IP multicast for about 20 years. However, the IP multicast solution has serious scaling and deployment limitations, and cannot be easily extended to provide more enhanced data services. There must be a long time for IP multicast to be deployed in internet widely. Recently, researchers have been attracted to deploy multicast service at application layer, these kind of solutions do not need to upgrade the underlay routers, and can support multicast as one need. In this thesis, we will discuss the key problems to deploy multicast service in internet for lager scale group communication applications.In order to support lager scale group communication applications in current networks, we present a hybrid wide area multicast service architecture. This architecture takes full advantage of the IP multicast's efficiency in local area and the application layer multicast's operability and practicability in wide area network. It can support large scale distributed simulation application, provide real-time and reliable data distribution services.When we construct the overlay network, adjacent nodes will be chosen as neighbors to decrease the delay in overly. Other applications also can benefit from an ability to predict round-trip time to other hosts without having to contact them first. A large number of hosts making independent and frequent measurements could have a severe impact on the Internet. We propose a hierarchy distance estimation method, which increases the measurement overhead a bit but provides more accurate distance estimation.Because the overlay network is a virtual topology constructed on top of the underlying unicast infrastructure, there are differences in network cost and routing constraints for overlay multicast and traditional IP multicast. The object of overlay multicast routing includes two aspects: one is to find a spanning tree that has minimum diameter, subject to degree constraint (MDDBST); the other is to accommodate the maximum number of sessions, with acceptable application performance. For collaborate applications such as DIS and Internet games, the diameter of the MDDBST spanning tree determines the speed of system. We propose greedy and heuristic genetic algorithms for MDDBST problem. In order to accommodate with the member dynamics, we design a distributed evolution strategy for application layer overlay multicast, which not only avoids the huge repetitive computations, but also shortens the diameter of the MDDBST spanning tree. To optimize the second object, we propose balanced degree allocation algorithm and loosen degree allocation algorithm that take into account the position and participate factors. The solution accommodates more multicast sessions and achieves higherbandwidth efficiency.Among the issues which delay the deployment of IP multicast, state scalability is one of the most critical ones. We propose a novel scheme, called AMBTS, which uses the concept of aggregated multicast, and employs tree splitting before aggregating. We design an algorithm to divide the leaf nodes of backbone into different sub-groups, and then splitting the native multicast spanning tree. We propose a scheme to assign a set of aggregated multicast trees to deliver packets for a group. Analyses and simulations show that AMBTS can greatly improve state scalability: the number of aggregated trees is bounded in a small fixed number, and the multicast routing entries in transit nodes decreases dramatically. AMBTS is a very cost-effective and promising direction for scalable transit domain multicast provision.In order to improve the state scalability of different multicast environment, we design multicast protocol ASSMBTS for Source-Specific Multicast and BEAMBTS for bi-directional shared tree multicast, which use the concept of AMBTS. These protocols are simple, easy to deploy and transparent to other domains.