Research On Reliable Multicast Protocol And It's Buffer Management

With the widespread use of multimedia conference, remote teaching, distributed interactive simulation, etc, in Internet, the research on large-scale reliable multicast protocols has become one of the hot topics in academia. Multicast is better suited for one-to-many and many-to-many data transmission than the traditional unicast and broadcast. Although the traditional IP multicast technology has reduced the demand of network bandwidth,solved the problem that the server load aggravate sharply along with the number increase in centralized conversation, but it does not guarantee that every member can receive the packets reliably and orderly in a multicast session. Therefore, the study on reliable multicast becomes particularly important. At present, buffer management in multicast is one of the key problems that restricts the multicast protocols further perfect and standardized. A bad buffer management mechanism may reduce multicast efficiency.This thesis firstly introduced the basic concepts of multicast, reliable multicast architecture and the related techniques of reliable multicast. Then, it discussed the randomized error recovery algorithm in randomized reliable multicast protocol (RRMP) in detail. RRMP distributed the error repair loads to group members, therefore avoided overload of the repair server and the packet loss due to the fault of repair server. RRMP used an innovative two-phase buffer management algorithm to optimize buffer management that obviously solved the problem of multicast packet delivery delay, and used randomized error recovery to achieve high robustness. Because RRMP sent retransmission requests by unicast and the randomized selection of long-term buffer nodes, it made the recovery delay increase with the increase of the group members. Aiming at the problems existing in the RRMP, this thesis put forward an improved reliable multicast protocol (IRMP) based on RRMP. IRMP divided group members into different domains. All domains constituted a hierarchical relationship. IRMP used the two-phase buffer management algorithm. That is, a short-buffer to meet the local recovery requirement and a long-buffer to meet remote recovery requirement. Local recovery used multicast to send retransmission request, and used the randomized back-off algorithm to reduce repeat messages. In remote recovery, a designated repair node was responsible for remote recovery and long data buffer. Simulation results in NS2 showed that IRMP had an effectively ability to reduce the buffer requirements and recovery latency, and also improved its scalability. Finally, the IRMP was applied in many-to-many reliable multicast. In order to adapt to the multi-source reliable multicast environment, this paper put forward message identification and a structure model of group members that were suitable for multi-source reliable multicast. Through the NS2 simulation platform, the experiment results proved the feasibility of the scheme.