Research On A Reliable Application Layer Multicast Protocol

Compared with IP multicast, Application Layer Multicast (ALM), which implementsmulticast function at end systems rather than network layer, avoids the restriction thatdeployment of multicast applications needs to change the network infrastructionand has muchmore flexibility and scalability. Besides, moving the network layer function to higher layerscan satisfy the desire for the richer services such as reliable distribution and sequenceddelivery. In the field of reliable multicast, application layer protocols are able to construct thedata delivery path along with group member management hierarchy, and enable applications todo error control and congestion control easily.The research work of this dissertation is mainly focused on the reliable application layermulticast. After investigating the theories and key technologies of ALM and reliable multicast,the author has designed an application layer multicast protocol for reliable data distribution,which is called Reliable Application Layer Multicast Protocol or RALMP for short. The majorcontributions of this dissertation are listed as follows:Firstly, a tree-based reliable multicast protocol is presented, in which the end-to-end delayis measured and is considered to be a factor influencing the cost of an edges in an overlaynetwork topology. The group member management hierarchy is in accordance with the datatopology that is organized as a tree. Parent node is responsible for packet replication,forwarding and retransmission, while child node is required to report packet loss event to itsparent so as to trigger loss recovery mechanism. At the same time, the related communicationamong hosts is restricted between the neighbors in a control topology. The efficacy of RALMPis demonstrated via a set of simulation experiments.Secondly, a self-organized distributed algorithm is proposed for the purpose that theoverlay network can be constructed and maintained. With this algorithm the tasks, such asjoining a group, exchanging information among hosts, measuring the states of links and soforth, can be performed automatically by the nodes. The algorithm can help evade thebottleneck of central server, which is usually encountered in centralized algorithms, andimprove the performance of RALMP. In addition, a fast patition recovery scheme and twotopology optimazing approaches are designed for RALMP to make the protocol more adaptedto the dynamic behaviour of the group members in a multicast model.Thirdly, RALMP makes use of TCP as a communication channel based on the underlyingunicast network. RALMP constructs the overlay network on the application layer, so the edgesof the topology are actually end-to-end logical paths over physical links. With the TCP pipes,these logical edges can be regarded as ideal data channels without transmission loss. In thisway, the error control mechanism can be concentrated on dealing with the loss of datatransmission interruption caused by the invalidation of ancestor host.Last, but not the least, to improve both network effciency and data reliability, a specialcontrol mechanism is adopted in RALMP. The mechanism is receiver-based, and combinedwith some features of sender-based mechanism. By this mechanism, the receiver detects thepacket loss and unicasts NAK to its parent, while parent can queries the status of its children.As far as error correction is comcerned, retransmission is done by parent who follows thepolicy of selective repeat. As a result, the nenwork bandwidth can be reasonally utilized, andthe data reliability can be guarenteed as well.