基於應用層群播之具回復性推拉式同儕式串流方法

Abstract

近年來，同儕網路的影音串流系統已愈來愈風行。影音串流的架構可以區分為基於樹狀架構與基於網狀架構。相對於網狀結構，樹狀架構會有比較低的啟動延遲，但是有部分節點失效時，其恢復力卻是很差，這是導致其傳輸率下降與接收端影音串流品質不穩定的重要因素。在本論文中，我們提出一個基於應用層群播之同儕網路多串流傳送機制稱為 HyStream，來改進以上的問題。首先，我們切割影音串流資料並建立多棵群播樹來傳送這些串流資料。此外，我們結合一個前向糾錯編碼演算法來回復遺失的資料。最後，我們結合拉式與推式方法，一旦資料遺失時，我們使用資料重傳的方法，即以拉式的方法來取回遺失的資料。模擬結果顯示，在不同的節點失效率下，我們的方法相對於SplitStream的遞送率有11.7%的改進。而在一個節點高度不穩定的環境下，我們的方法的遞送率比 CoolStreaming 高了2.2%。相對於 CoolStreaming，HyStream 中百分之九十的節點的啟動時間減少了35秒。相對於 CoolStreaming 與 SplitStream，我們的方法需6% 額外的封包。而對於 SplitStream 與 CoolStreaming，我們的方法大約多了0.5%的控制訊息流量。

P2P streaming systems are getting more and more popular in recent years. The streaming architectures can be classified into tree-based and mesh-based. The tree-based architecture has low start-up delay, but has less resilient to node failures compared to the mesh-based architecture, and it would result in a low delivery ratio and instable quality of received multimedia. In this thesis, we propose a P2P multi-streaming scheme called HyStream based on application level multicast to improve these problems. First, we split video streaming data and build multiple trees to transfer streaming data. Second, we integrate a forward error correction (FEC) algorithm to recover lost data. Finally, we combine the pull method with the tree-based architecture, which is based on a push method. When encountering data loss, we use a pull-based data retransmission method to retrieve lost data. Simulation results show that in average our approach has 11.7% improvement in delivery ratio against SplitStream under various node failure rates. The delivery ratio of the proposed HyStream is 2.2% higher than that of CoolStreaming in a peer churn environment. The start-up delay of 90th percentile nodes of HyStream is 35 seconds shorter than that of CoolStreaming. Our approach has low overhead of 6% extra packets compared to SplitStream and CoolStreaming. And the extra control overhead is not more than 0.5% even in a high peer churn environment compared to those of SplitStream and CoolStreaming.