在點對點串流系統的合作式資料排程演算法

Abstract

本論文提出在點對點串流系統中傳送端資料傳送排程演算法。點對 點串流系統在最近相當熱門，因為沒有伺服器成為瓶頸，過去幾年成 為主流研究的課題之一。對於資料排程演算法，也有相當多不錯的方 式被提出。不過大部分的研究都主要集中於接收端的排程演算法，對 於傳送端的描述只有講個大概，並沒有具體的演算法發表。有鑑於傳 送端資料排程演算法的缺乏，所以本論文在觀察點對點串流系統特性 後，研發出具體描述之傳送端演算法來填補此領域的不足。相信藉由 傳送端和接收端雙方資料排程演算法的合作，能使整體系統的效能和 結果能有所提升。

Media packet scheduling is an important ingredient for peer-to-peer (P2P) streaming system. Most of the previous work focus on how to efficiently pull desired media chunks from neighbors in order to achieve their designed goal. To best of our knowledge, there is a lack of studies of how a peer serves multiple requests sent by its neighbor under limited upload capacity. In this work, we proposed a sender-side scheduling in order to cooperate with receiver-side scheduling. In this work, we first identify that the difficulty and challenge of sender-side scheduling. By observing the system behavior about utility of upload capacity and the reason of content bottleneck, we proposed a sender-side scheduling in order to maximum the utility of each peer upload capacity and minimum the probability of content bottleneck happen. We approach these two goals by using utility-oriented requester decision and diversity-oriented packet decision. Utility-oriented requester decision is the algorithm that which peer sender should serve such that the minimum utility of peer upload capacity is be maximized. Diversityoriented packet decision is the method that when served peer decided, which packet should send to this peer for the sake of the increasing the system diversity which implies that the happen of content bottleneck is be minimized. Simulation result shows that our proposed scheduling is better than best effort sending packet on constant bit rate and variable bit rate environment. In addition, we shows that proposed sender side scheduling can improve the performance of receiver-side data scheduling. Simulation result shows using proposed sender-side scheduling cooperate with receiver-side scheduling, user perceptual quality can be enhance about 1.83dB for well-designed receiverside scheduling, and enhance 7.65dB for simple receiver-side scheduling. ix