隨選視訊的點對點儲存策略

Abstract

多媒體影音服務，由於影音串流的資料量大、頻寬需求高、持續時間長，傳統式IPTV Client/Server模式多以CDN架構利用multicast串流傳輸方式來服務各個客戶端，來減輕server 運算負載及解決頻寬不足問題，但客戶端所接受到的影音資料有一定時段範圍。 P2P(peer to peer) network傳遞影音串流的資料，藉由peer間彼此分享、互相傳遞，來避免server 運算負載及頻寬不足的情況，並且系統可以有較佳的擴充性及容錯能力，但影音串流資料量與時間成正比，雖然各個peer皆可以分享暫存資料，但隨著時間越久、peer的加入和離開，而導致影音串流資料雜散在網路中，衍生出新加入的peer搜尋來源資料困難及延遲等相關問題。 我們提出一種結合centralized Server、動態的super-peer以及較無計算能力的peer為基礎的P2P網路架構，來進行live video streaming及VOD(Video on demand)的影音資料及應用訊息交換，並將瀏覽過的影音資料進行暫存的動作，進而達到影音資料分享的功能。我們提出adaptive replacement cache的演算法並與兩種傳統暫存的方法，No replacement cache與continuous replacement cache，作比較。並以熱門程度變換、暫存時間及相關的參數進行模擬測試，比較這三種方法在Peer進行影音搜尋及分享的平均資料量。

For the multi-media services, video streaming application is charactered by its large data, high bandwidth and long duration playback. To reduce streaming server operation load and insufficient bandwidth problems, most of traditional IPTV Client-Server models use multicast streaming based on the CDN architecture. However, the video and audio information that can be received by clients is limited to a certain time frame. Streaming applications in P2P networks are more and more popular recently. By adopting data sharing and information exchange techniques to resolve the server operation load and bandwidth problems, P2P network system has better scalability and fault-tolerant ability. However, the amount of data produced by video streaming server is directly proportional to the time that service has been provided. Although each peer can share its data, with time going by and the peers joining and departure, the video information distributed among peers become disordered. This results in more difficulties and delay time for the peers to search for the sources of video information. We propose a p2p network architecture which consists of centralized Server、dynamic super-peers and low computing power peers. When doing the live video streaming and video on demand applications, peers will cache video data that they have viewed, then share and exchange its cached information with other peers. We propose a technique, called Adaptive replacement cache, and compare it with two traditional cache methods, called No replacement cache and Continuous replacement cache. By changing segment popularity, cache size and the related parameters on our simulation test, comparison among these three methods was conducted for the average amount of video data shared by the peers in the P2P network.