Research On Multicast Communication Techniques For Large-scale Network

When Internet is widespread rapidly, the streaming media services attract more and more Internet users. Unfortunately, the traditional communication models couldn't satisfy the need of one-to-many data distribution in large-scale networks. In this paper, we provide a new group communication model named MixCast, which combines unicast and multicast communication model.The core of MixCast is to use unicast communication between the domains and use multicast communication inside the domains. Furthermore,MixCast constructs a data forwarding tree on application level to distribute streaming media data from source to all receivers.Through the constrastive experiment in unicast, IP multicast and MixCast, we prove that MixCast is better than unicast in average delay,packet lose rate and network throughput, and easier than IP multicast in deployment. In MixCast model, we focus on the problems of tree construction, expansibility, reliability and heterogeneity.To solve the problem of application level multicast tree construction, we provide a new protocol named MTP, which includes two distributed algorithms — MTP1 and MTP2.We also provide a centralized algorithm named OST and compare the 3 algorithms with OMNI. In experiment results, we could find that MTP2 is better than MTP1 in average delay and MTP1 is better than MTP2 in average join time and join control overhead. The centralized algorithms are worse than the distributed algorithms in expansibility, although they are better in average delay.To promote the expansibility of MixCast, some strategies are adopted, including using distributed algorithm on forwarding tree construction, combining unicast and multicast communication models, decreasing the average delay of transferring streaming media, reducing the control overhead of protocol, and etc.To enhance the reliability of application level multicast forwarding tree, we provide a pre-computation algorithm — PCP, which is derived from the idea of resource reservation mechanism in QoS. In PCP, each node in the forwarding tree pre-computes a backup parent node.When its normal parent is disabled, the node could receive streaming data from its backup parent node immediately. In experiments between PCP and another pre-computation algorithm ROT, we could find that PCP is better than ROT in average join time and average join overhead, but is worse in backup link delay.The traditional application level multicast protocols always ignore the problem of heterogeneity, which will bring the instability of forwarding tree.In this paper, we provide a...