應用層多播網路之即時影音串流設計

Abstract

近年來使用者對多媒體的需求帶動播放，擷取及傳輸設備快速普及，加上各種寬頻網路的佈建，使得隨時隨地透過網路與他人分享即時影音成為可能。本論文著重於設計一個可讓使用者建立頻道分享即時擷取的影音，並可在多個頻道間切換收看的平台。 我們參考多種即時影像於網路傳輸的模式，IP Multicast普遍被網際網路提供商用來提供triple play服務(即電話、電視及網路三項服務共構)，但其需要特殊的多播路由器的支援，對業者來說建置系統的門檻也相對高出許多。許多研究都把重心移到應用層的多播，利用上層資料結構記錄群組成員及傳送路徑。目前即時串流的應用層多播解決方式分為雙層架構與點對點架構。其最大的共同點是由來源端建立起一棵多播樹並且擔任管理的工作。 本篇論文提出一些改良方法。我們以混合式點對點網路架構建立的應用層多播網路將即時影音傳輸至各個使用者。頻道中的使用者被組織成一棵多播樹，每個節點記錄此傳播樹上的一些成員節點，當加入/離開頻道或其父節點離開時，都由節點自行處理，此舉可減少來源端的負荷，並且將系統負荷平均分散到使用者身上。此外，當使用者暫時離開頻道時，這些成員節點的記錄會暫存下來用以縮短重新加入此頻道所需花費的時間。

In the recent years, the need of multimedia services promotes appliances, such as MP3 player, digital camera, and digital video, become more and more popular. On the other hand, the distribution of many types of access networks makes it possible that users can share their live video streaming at any time and any where. This thesis focuses on designing a platform where users can create channels to share their streaming and switch between these channels to watch the shared streaming. We studied many real-time streaming delivering solutions, and the IP Multicast is proposed by many ISPs to construct triple-play service (the combination of the television, telephone, and high-speed internet access services). However, the IP Multicast needs the supports of special network equipment – multicast routers. This increase the building cost for the service providers. Today, most of research in multicast has been moved to application layer. The group members and delivery path are kept in the application layer. The solutions classified into two-tier and peer-to-peer architecture. The common point of these solutions is that the source node builds a multicast tree and serves as a manager of the tree. In the thesis, we propose a refined method. We construct an application layer multicast tree based on a hybrid peer-to-peer model to deliver streaming content to every user. The users in a channel are organized as a multicast tree and every node keeps information of some member nodes on the multicast tree. When a node joins/leaves a channel or when its parent node leaves, the node can use the information to handle these events. Therefore, the burden of a channel can be dispersed to every participant. Moreover, the kept information is used to reduce the rejoining latency after a user departs a channel temporarily and switches back.