在MOSP環境中研究多媒體串流的方法與實作

Abstract

MOSP（Mesh Object Service Protocol）是一個開發中的計劃，目標是建立一個新的分散式網路協定與環境，並以物件導向的結構來設計。使用者在連上MOSP網路後，可以透過MOSP協定來存取網路上的各個服務。在其上使用多媒體服務的時候，需要即時地傳送龐大的影音資料，可以透過串流的技術來達成。而網路中串流的主要標準是Real-time Transport Protocol（RTP），選擇協定的考量有：可靠度（TCP）和即時性（UDP）。UDP在串流服務上提供許多便利性，但也帶來其他的限制，加上MOSP是架構在TCP之上的協定，因此在本篇論文中，我們要研究TCP上的串流服務之建置與效率。由於TCP是可靠傳輸，其對資料傳送的保証還有他的流量與壅塞控制機制，會合理的利用頻寬，但是卻帶來不可避免的延遲，這在即時多媒體應用上是很大的缺點。 本論文中提出一套smart server / simple client的機制，利用從使用者端及TCP層所得到的資訊，在伺服器端主動增加或減少資料量，以符合目前頻寬狀況，進而減低延遲。我們所參考的參數有傳輸所花費時間、目前網路頻寬、平均網路頻寬、目前媒體資料產生率、平均資料產生率、目前伺服器中累積資料數及資料大小，這些都會用來當作傳輸決策的參考，在有限頻寬中最大化傳輸量及降低延遲。

MOSP (Mesh Object Service Protocol) is a developing project for establishing a new distri- buted network protocol and environment, which supports object-oriented structure. After success- fully connecting on MOSP network, users can access different services on remote servers over In- ternet via MOSP protocol. Large amount of data transmission is needed for multimedia applications, in which case we use streaming technique to accomplish the goal for real-time transmission. A primary adopted standard of streaming is RTP, Real-time Transport Protocol, which can be built on top of UDP or TCP protocol. UDP protocol provides convenience in timely transmission with many other restricts like additional awareness of congestion control and firewall configuration. With these restrictions plus MOSP be a TCP-based protocol, we’re trying to establish a TCP-based streaming technique efficiently, under drawbacks of TCP in case of delay for real-time applications. In this thesis, we proposed a smart server / simple client architecture, reducing or increasing data transmission rate accordingly on server side with feedbacks from clients or TCP layer, in order to reduce the amount of delay to fit in various network bandwidth conditions. Parameters we put into consideration for transmission decision include the delay of each transmission, current avai- lable bandwidth, average bandwidth, current source data rate, average source data rate, current data amount and size accumulated in server buffer, for maximizing the amount of data transmission and minimizing the delay and latency of streaming service, under limited network bandwidth.