在目前的網際網路上，使用群播傳輸技術來傳送資訊，不能保證接收者都可以正確無誤地接收到所有資訊。視訊會議等影音應用服務並不要求播放的內容需被百分之百地傳送；然而在金融業務、軟體更新以及醫療影像教學等網路應用上，可靠群播是必需的。最近幾年，可靠群播的議題引起了熱烈的討論，也有人提出了不少的方案。然而，在這些可靠群播協定中，大多數仍存有可擴充性和不支援即時傳送的問題。 在本篇論文中，我們提出一個新的可靠群播協定－具多重樹狀結構之可靠性群播傳輸(MTRM)，目的為在異質網路環境中(包括對稱和非對稱性網路)提供可靠和適時的群播傳輸。為了在非對稱網路中，例如纜線數據機、非對稱數位用戶迴路，達到可靠傳送多媒體資訊，我們在每個會議(session)中，不僅保有原來的群播群體(group)還新增了一些具有修復能力的群體。每個修復群體的根源為一特別的伺服器(縮寫為指定伺服器DS)，負責恢復在它服務領域之內的成員們所失去的資料。 此外，MTRM採用接收端自行偵查錯誤、混合NACK-FEC技術、階層式樹狀結構以達到即時性。最後的分析結果顯示出MTRM不僅在效能上有所改善，而且大大地減少了發送端的負擔和資料的傳送時間。 Distribution of information using the multicast in current public network does not guarantee reliable delivery. Reliable multicast is required in applications such as financial transfer, software updates, multimedia delivery, and medical images. Recently there has been a great interest in reliable multicast for transmitting the same information to a group of receivers simultaneously. Unfortunately, most of the existing reliable multicast schemes have scalability problems and do not support real-time traffic. In this thesis, we present a new reliable multicast transport protocol, called Multiple Tree-based Reliable Multicast protocol (MTRM) for reliable and timely delivery of multicast traffic using IP multicast in a heterogeneous network (included symmetric network and asymmetric network) environment. For each multimedia session, the MTRM protocol maintains a single multicast group and several repair groups, to work in asymmetric network, such as cable modem, or ADSL carriers. A novel special server, Designated Server abbreviated to DS, works as the root of the repair group to be responsible for recovering lost data within its service domain. In addition, the MTRM is receiver-initiated error detection, hybrid NACK-FEC technique for reliability and based on a hierarchical tree-based structure in which the DS groups the receivers in local regions to avoid feedback implosion. Performance results show that MTRM not only improves throughput and usage of resources but also significantly reduces the processing burden of the sender and end-to-end latency.
