Research On Multicast Reliable Transport Control Mechanisms In Wide Area Networks

With the rapid growth of the Internet, there are being of more and more new applications, such as large-scale network audio/video conference, push technique, publish/subscribe, large-scale collaboration computing, IPTV, large-scale longdistance learning system, distributed grid computing, share white board, and other, such as user group software update, update for mirror and high speed cache, Distributed Interactive Simulation, and so on. All these posses characteristic that send same message from one to many or many to many. This may happen on a hundred thousand hosts at the same time in the Internet. This all needs higher bandwidth. It goes beyond much more the capacity of unicast. Substantive bandwidth will be consumed by using unicas to distribute message from one to many or many to many. It is challenge to Internet service providers how to provide effect and steady service based on network resource in existence. IP multicast is an efficient way for one to many or many to many packets distribution at the network level. It has become key support technique for multicast application.There are many active research areas in multicast. One is research on multicast reliability. In this dissertation, it is about research of multicast reliability, including IP multicast reliability, application level multicast reliability, and congestion control. The contribution of the dissertation can be summarized as following:First, aiming to problem of scalability in reliable multicast, i.e., along with the increase, protocol performance is poor. Therefore, we provide a reliable mutlciast mechanism based on gossip, and provide three layers model of reliable multicast. Basic idea is that loss recovery is divided two layers, probabilistic reliable layer and determinate reliable layer. After probabilistic recovery, most loss message hosts get recovery, and determinate reliable layer only deals with a few loss recovery. Feedback implosion can be controled. Our simulation results show that our protocol performance is improvement from the viewpoints of scalability.Second, in ALM, multicast related feature, such as group membership management, multicast routing and packet replication, are implemented at end- host instead of routers. A multicast distribution tree is constructed in the application layer, so all nodes in this tree are end-hosts. However, ALM has a serious problem that the multicast distribution tree is intrinsically fragile and an end-hose failure causes tree partitions. To deal with problem, we propose based on k-regular graph tree construction protocol that constructs a robuse distribution tree against end-hose failure. Simulations prove the effectiveness of the proposed control mechanism.Third, one of the significant challenges associated with multicast delivery is providing congestion control mechanism.Layer multicast for addressing receiver heterogeneity has been provided. In layer multicast, receivers subscribe and unsubscribe to layers according to network congestion condition. It is important to how to measure network. We propose a based on MINC (multicast inference of network characteristics) network monitor, and it can implement dynamically subscribe layer multicast and improve information distribute quality. Simulation result show effective.