| IP-multicast is based on an open, best-effort service model. It does not provide QoS or delivery guarantee. In order to ensure the reliability of transfers, a lot of reliability multicast protocols have been proposed. However, none of them is able to satisfy all applications. Up to now, each reliability multicast protocol is designed only to satisfy some type of applications, and no one becomes a standard protocol.This dissertation analyzes some popular reliability multicast protocols, which are designed for the applications that may have millions of nodes. In order to improve their scalability, these protocols usually use some complex technologies, and can not validly support the file transfer applications with multiple recipients in a telescience network. In this context, a reliability multicast file transfer protocol of application layer, Simple Multicast File Transfer Protocol, is designed to suit the ground network architecture and the characteristics of space experiment data distribution in a telescience system.In order to reduce the sensitivity of network delay and avoid the network congestion, SMFTP employs the transfer model that divides a file into one or more blocks and transmits its data in passes. Out of the security concern, SMFTP takes the close group membership management solution, which can facilitate the extension of SMFTP in the future. SMFTP adopts a mixed mechanism combining sender-reliable and receiver-reliable error recovery means. The receiver-reliable error recovery approach is used to report transfer errors, and the send-reliable error recovery approach is used to inform the sender which buffer space can be freed.Finally, the approach of analytical evaluation is used to determine the maximum throughputs of SMFTP, and it is compared with the other type protocols'. In general, sender is the throughputs bottleneck of reliability multicast protocols, and the conclusion is that when the number of receivers is not very large, say several hundreds, the throughputs of sender side of SMFTP are higher than other protocols, and the possibility of NACK implosion on the network near sender side is relatively less. |