應用層多播線上即時串流

Abstract

IP Layer Multicast提出解決網路路由器(router)封包重複於路由之間傳送問題的方法，使靠近客戶端的路由器可以針對特定群組而送出封包，無需經由來源端到客戶端當中的所有路由器均參與其中。因此大量減少網路上不必要的封包傳送，使得一對多服務變成容易許多。然而由於IP Layer Multicast背後的問題諸如路由器管理、缺乏區域路由器之間多播協定、拓展能力、佈建、與異質網路結合、及對上面協定層功能的支援如流量及擁塞控制等，因而使得至今仍未被具體實現。 目前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程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傳播樹並且擔任管理的工作，在(1)指派服務節點；(2)節點加入與離開；(3)中間節點突然遺失時傳播樹重建都需由來源端負擔。在此論文中我們提出一些改良的方法，由每個節點記錄一些服務節點並定期更新，在加入/離開系統及當服務節點遺失的時後由使用者自行處理，如此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，在傳播樹的建置中，我們會考慮節點到來源端間所經過的Hop及數傳輸時間，使得從來源到每個節點間所經過的Hop數與時間都能達到儘量最少，以便縮短傳輸延遲現象。讓每個節點擁有服務其它節點的能力，而來源端僅需提供多媒體串流資源，故在這系統上的每個節點都有能力成為來源端而佈建自己的傳播樹。

IP layer multicast provides a solution for the problem of duplicated packets retransmission between routers, it makes only the router nearby the clients to duplicate packets to specific groups, not each router between sources and clients take part in the delivery path. Therefore, that can reduce the unnecessary packets transmission over the Internet, makes it easy for one-to-many delivery. However, many problems behind the IP layer multicast such as routers management, lack of a robust multicast routing protocol between local (inter-domain) routers, scalability, deployment, combination with heterogeneous network, and support for higher layers functionality such as flow and congestion control. As a result, IP layer multicast has not been concrete accomplished up to the present. Today most of the research in multicast has been moved from the IP layer to the application layer, group members and delivery path are kept in the data structure in application layer. We can summarize the solution for live streaming using application layer multicast into two-tier and peer-to-peer architecture. The major common point of these are the source build a distribution tree and serve as the manager of the system in (1) assigning the service node, (2) peers joining and leaving, (3) node failure and tree reconstruction. In this thesis, we propose a refined method to diminish the burden of the source and distribute the system load balance to the users. Each peer takes down some peer’s information when joining to the system, updating these information periodically, processing node failures. At the same time, we will consider the numbers of hops and transmission time from each peer to the source to make sure the total hops between each peer and the source is best-effort least, such that the delay can be diminished. In consequence, each peer has capability of serving, and the source just provides media stream resource such that each one has the capability to be the source and build its own distribution tree.