基於同儕式網路隨選視訊串流的高強度區塊作種與排程演算法

Abstract

在基於同儕網路的隨選視訊系統中，隨選視訊伺服器端提供影音內容，客戶端則以點對點方式下載並分享影音內容。由於檔案分佈的低同步性，加上快取空間有限，導致擁有的影像片段相異，造成同儕網路的分享與下載效能不彰。另外考量到網路使用花費與網路頻寬限制，隨選視訊系統客戶端必須協助伺服器端作種（Seeding）分享以節省其網路租用成本。如何解決這些問題並增進分享效率是本論文的研究課題。 針對此問題，我們將影片切割成若干個小型的影音區塊（Segment）作為下載與分享使用的單位，並分別製成種子檔案（Torrent）與作種，進而根據影像播放的優先順序，對作種與下載的任務設計出一套排程方法，並預測下載速度與播放緊急程度，達到避免播放影片時產生的中斷情形，提升影片播放品質，並增進同儕間之作種分享效能。 由實驗中我們觀察到下載演算法中緊急指標之低水線越低，對使用者播放中斷情形有較佳之影響，且作種方面以高頻寬配合較多作種個數會表現最佳之結果。

In P2P VoD system, VoD server supports the video content and clients download and share by P2P. Further, VoD clients have less synchrony, cause them to share weakly. Consider the cost of network and limit on bandwidth, clients must help sharing server to reduce costs. The approach of this thesis will solve these issues. We separate a film into many small video “segment” as a unit of downloading and seeding, take into torrent files and to be seed. Based on the time priority of playing file, we provide a segment scheduling algorithm, expect enhance smoothness when user watching files, and have outstanding performance of sharing among peers. In the experiments, we found out when L.U. is low, user experience is better. Seeding algorithm have the best performance when high bandwidth with large seeding size.