限制延遲時間的點對點即時影音串流系統

Abstract

內容散佈服務(Content distribution services, CDSs)已成為今日網際網路的主要流量來源，而根據統計，其中一半的流量來自影音串流服務。即時影音串流(Live streaming, LS)，提供人們一個平台用來即時分享身邊正在發生的事件，將成為未來主要的影音服務之一。Playback delay，即事件發生後相隔多久時間才能被其他使用LS服務的使用者觀看到，是評量一個LS系統最重要的依據，特別是當提供HD品質等高頻寬需求的LS服務時。因此，本篇文章專注於研究如何在樹狀P2P網路中提供LS服務，使得系統能夠在限制Playback delay上限值與提供高串流資料率時服務更多的使用者。我們提出了兩個演算法的組合，PCL生成演算法及區域最佳化演算法。PCL生成演算法讓新加入網路的節點能夠被分配到適當的樹狀網路位置，使得能貢獻較多上載頻寬的節點能夠經歷到較小的Playback delay。而區域最佳化演算法則提供系統一個分散式的方式來優化系統網路，用以降低系統平均Playback delay。模擬結果顯示，相較於僅考慮Playback delay的演算法，我們所提的演算法組合能服務較多的使用者，尤其是提供高串流資料率與限制Playback delay上限時表現尤其突出。除了模擬工作以外，我們實作了一個基於樹狀P2P網路的LS系統雛型。該系統雛型讓我們能以真實網路環境來驗證與改進所提演算法，並已實際被應用在教學用途上。

Nowadays, content distribution services comprise most of the Internet traffic, and about half of them is contributed by video streaming. Live streaming (LS), which provides people a way to share what are currently happening in the world, will be one of the main streams of video services on the Internet. Playback delay, the time interval between events occur and the event video are watched by LS users, is the most important metric of LS systems. In this paper, we focus on providing LS services in tree-based P2P networks, so that systems could serve more users under high bandwidth requirement and maximum playback delay constraint. We proposed two algorithms: PCL algorithm, and local optimization algorithm. PCL places newly joining peers in the tree properly so that peers which contribute more upload bandwidth could experience lower playback delay, and local optimization aims to lower the average system playback delay in distributed manner. The simulation result indicates that our algorithms allow systems to serve more users than the system which adopts an algorithm that only considers playback delay. In addition to the simulation work, we have implemented a tree-based P2P LS prototype, which allows us to verify/improve the future proposed algorithms in real Internet.