Probabilistic Multicast Trees

Abstract

Delivery of the same data content to many clients simultaneously over the Internet continues to be a challenging problem. Multicasting using a single tree structure for data distribution has been shown to be an effective methodology for distribution of data. Using the tree structure to distribute data relieves the source node from the burden of trying to unicast to each client and is efficient because the data delivery burden is distributed over all the participating client nodes. Using multiple tree multicasting further distributes the transmission burden over more participating client nodes and it improves the efficiency of the data distribution. Multiple multicast trees can also be used to manage dynamic behavior of the underlying network. We introduce a methodology which improves data delivery latency and efficiency upon current multiple tree multicast methods. This methodology incorporates a feedback mechanism, randomness and a weighted tree selection mechanism to determine the most efficient multicast tree for multicasting