Research Of Congestion Control In Clustered Video Server

With the rapid increasing demand for the video service over Internet such as Video on Demand, video server are used in many fields and the clustered video server has received more and more attention because of its good scalability, reliability and cost/efficiency. To provide the high quality video service, the high bandwidth is required. However, nowadays Internet provides best-effort service and the network bandwidth is time-varying. When the network is congested, there will be a lot of packets to be dropped by the network. In order to guarantee the quality of service, the server must perform the congestion control to reduce the packet loss ratio. Unfortunately, the currently existing congestion control schemes cannot meet such requirement perfectly. For example, some schemes need lots of computation, which result in the limited number of the concurrent streams. And some schemes need the special support of the multimedia players and cannot be applied to the popular multimedia players like QuickTime player. Some schemes lose the generality since they are only suitable for the files encoded by some special coding methods. So a novel congestion control scheme is proposed in this thesis to improve the quality of service. According to the characteristics of the single stream based transmission of video sequences coded with MPEG (Moving Picture Expert Group), the international video coding standard, which is used in the Video on Demand system, the following methods are employed in this proposed scheme. Through smoothing the bit rate of a decode unit at the system layer of MPEG, the rate smoothing mechanism reduces the buffer size in the client and shortens the start latency. In the rate control scheme, we utilize AIMD (Additive Increase and Multiplicative Decrease) adjustment algorithm and consider the rate provided by the special media file, thus can make the client recover from the congestion quickly. And in the frame dropping scheme, firstly the transmission priority of the multimedia files are classified according to the number and type of the dropped frame, so it can quickly determine which frame should be dropped and greatly reduce the computational complexity.This proposed scheme has been applied to the WanLian system, a clustered video server successfully. The experimental results show that after the proposed scheme is applied the loss rate is less than the before and the audio interruption and image mosaic rarely occur when the network is congested.