Modeling Of Variable-Bit-Rate Video And Improvement Of Streaming Server

The variable-bit-rate (VBR) stream has made up a main part in the nowadays broadband traffic. Due to its better utilization of band width and better quality, VBR video has been widely used in video-on-demand systems, video surveillance systems and video conference systems.The vibration of data rate of VBR video may put traffic bursts on networks. It challenges the design of streaming servers on properly allocating resources, that is, to guarantee the QoS while minimize the waste and applying congestion control. These problems cannot be solved until a proper model of VBR video is found.Researches have proved the existence of both short-range dependence and long-range dependence in VBR sequence. While the traditional time series analysis deals only with the former, the modeling of the latter should employ self-similarity. This idea was used to describe the similarity of statistical features under different time view. Self-similarity has been found in quite a few fields of computer science.As the existence of long-range dependence and self-similarity in VBR sequence, traditional time series analysis may not be useful. The paper applies a fractional ARIMA (F-ARIMA) model on the VBR video to describe both the short-range and long-range dependence, and makes a thorough analysis on empirical video sequences.The paper also introduces the construction of a streaming server, which acts as a proxy between live video sources and users. It is used in video surveillance systems and video conference system. The server includes a data acceptor, a data dispatcher and a client acceptor. Patterns like reactor, acceptor-connector, active object and non-blocking buffered I/O are utilized to guarantee efficiency, maintainability and reusability. A buffer is proved to be essential to shape bursty data and traffic. To achieve a good design, we also make queuing analysis under the self-similar input. The self-similar parameters of two empirical video sequences are used to analyze the queue requirement of the server. The QoS of different applications and hardware provided is compared. Finally, we provide the architectural designs under several applications.