| Summary: | 碩士 === 輔仁大學 === 資訊工程學系碩士班 === 107 === With the advancement of cloud technology and network transmission speed, mobile phones have become very popular. Many audio and video platforms have gradually risen, resulting in a large increase in network traffic. The bottlenecks of high cost and low scalability of the client–server model makes it difficult to serve a large number of users at the same time. Therefore, many audio and video services began to operate under the peer-to-peer architecture. The stability of the peer-to-peer architecture is criticized by the public. Users consider the system stability as the most important attribute of a live streaming system, so most peer-to-peer live streaming systems often sacrifice delay and cost in order to achieve high stability. The focus of live streaming is the real-time interaction between the viewers and the live broadcaster or the real-time information of the live content. Therefore, we consider cost, stability and delay are all indispensable indicators.
In this thesis, we propose a dynamic topology maintenance strategy. On a tree topology to control the system architecture and reduce transmission costs. Moreover, we also use the buffer overlap decision mechanism to reduce the delay of live streaming, and we use peer grading strategy to improve system stability and reduce the impact when an upstream peer gets offline. In addition, for each peer we assigned a backup peer to provide audio and video clips for that peer when it lost the upstream transmission peer. In this way, we enable each peer to receive audio and video clips stably, reduce the delay between peers and servers, and improve the smoothness of live video playback. The experimental data show that our proposed strategy can improve the streaming quality by 2%~5%, and the delay by 15%~20%. In comparison to the mesh topology, the tree topology does not need to send messages to neighbors frequently, so the transmission cost is lower. This strategy can improve the performance of the overall live streaming systems.
|
|---|
