改善起始延遲及畫面品質之使用者經驗於使用可調式視訊編碼的點對點串流系統

Abstract

傳統主從式(Client/Server)架構的串流技術易造成供應端負載過大、供應源的頻寬不足而使服務效能低落；而點對點 (Peer-to-Peer) 架構則藉由分散供應源以減輕主伺服器之負擔，大為改善主從式架構所遭遇的問題，因此點對點串流系統越來越風行。但點對點串流技術的相關研究大多著重於挑選最佳下載節點、節點分布最佳化、群組建構等點對點通訊協定之演算法，少有文獻特別針對改善使用者經驗(Quality of Experience，QoE)之方法討論，像縮短等待時間及提升觀看畫質等。當系統效能再高若使用者經驗過差也無法提升其使用率。因此對於串流系統，使用者觀看時的感受也應該被重視。同時，不同使用者的使用需求也不甚相同，因此我們在點對點網路環境下利用可調式視訊編碼技術(Scalable Video Coding)的分層影像品質特性，改善起始延遲以及畫面品質之使用者經驗，讓使用者在觀看串流影像時有最小的延遲時間並能享受到最佳的畫面品質。 本論文提出一Layer Deadline Scheduling(LDS)演算法，藉由計算各階層的最終播放時間排程下載順序。若實際下載時間與預測衝突時，便以優先權值決定下載或放棄。實驗結果顯示，比較四種演算法：(1) LDS Algorithm、(2) Random Selection Algorithm、(3) Smooth Algorithm without delay、(4) Smooth Algorithm with 1 sec delay，當網路環境的RTT/2從0 ms增加到300 ms時，LDS演算法表現最佳。畫面品質更是Smooth Algorithm的兩倍以上，而有效的下載率更可達80%以上。

Client/Server streaming is susceptible to server’s overloading and liable to exhausting the usage of server’s bandwidth; hence, server’s service is degraded. The above deficiencies can be solved by P2P streaming where a resource provider is split and distributed to multiple nodes over a network, so the server’s loading can be mitigated. Therefore, P2P streaming becomes prevailing in recent years. Most of the researchers focus on peer selection, network structure, group organization, and so on. The issue of improving Quality of Experience (QoE) which is used to specify user’s perception is scarcely discussed. User waiting time and picture quality are two of the metrics for QoE. If QoE is low, no matter how system performance is, the usage rate cannot rise. Therefore, the metric of QoE must be taken into account by a streaming provider. But different users have different usage demands, the layered quality characteristic of Scalable Video Coding (SVC) are adopted to enhance the QoE. This will lead to the best picture quality and the least user wait time for users. In this study, we propose a Layer Deadline Scheduling (LDS) Algorithm to schedule all layers by their playback deadline and calculate the current available bandwidth to make sure if the streaming file being downloaded can be received in time. If not, whether to download a streaming file or skip it depends on its priority. A comparison of the empirical results of four algorithms: (1) LDS Algorithm, (2) Random Selection Algorithm, (3) Smooth Algorithm without delay, and (4) Smooth Algorithm with 1 sec delay, shows when RTT/2 ranges from 0 to 300 ms, LDS Algorithm has the best QoE. The picture quality with LDS Algorithm doubles than that with Smooth Algorithm. Moreover, the proportion of discarded layers with LDS Algorithm is as low as 10% of all files.