| Development of multimedia technology allows users to experience more and more services of the information society. With network bandwidth constantly speedup, the continuous development of video coding technology, innovation of multimedia server, enterprises and individuals can have a better experience of multimedia from the network. The mainstream of current coding techniques are used stream of variable bit rate (VBR), but the current media server technology research for variable bit rate video traffic prediction has some limitations, especially for the long-range correlation. This will lead to sharp fluctuations of streaming media server, dropping frames seriousl and affecting the quality of service. Current streaming media server are not designed by different parameters of array according to different media files, this will result in additional costs, reducing server performance. The I/O scheduling and convergence strategy are not optimize the bit rate according to the characteristics of streaming media. We study streaming media server optimized for VBR video in the following three aspects.Firstly, we introduce fractional autoregressive moving average model to model multimedia video traffic to analyze the next period traffic data, especially for the VBR video stream of short-range correlation and long-range correlation model. In fact, VBR video stream associated with good short-range and long-range correlation. The traditional time series analysis describe the short-range correlation, and describing the long-range correlation is necessary to introduce self-similar, because there is self-similar VBR video stream and long-range correlation, time series analysis can not be correctly modeled, so the use of FARIMA model (autoregressive moving average model) to describe the long-range correlation. Leads us through the ARMA model to study FARIMA model, with an emphasis on FARIMA (p, d, q) and Hurst parameters were studied. By our definition of the parameter estimate, and validate the model, if the time inteval is 1/24 seconds, the error of next unit of time flow forecast is less than 1%, good results obtained.Striping-based allocation of Mage-MS was introduced to allocate disk space. We allocate large disk space to fixed-size Mage. Mage-based allocation method is somewhat similar to previous studies, this method and traditional methods is a fundamental difference between a fixed-size of the Mage. More importantly, when client retriev streaming media, only the required data can be prefetched into memory, rather than the entire Mage, these data are on order of distribution on the disk surface. Another advantage of Mage-based allocation method is to estimate the cost during reading. Because the size of a stream request size in each round never exceeds Mage, the access cycle, up to serve only two Mage request the same stream, which avoid the shortcomings of previous allocation methods - large amounts of data replication. Experiments show that VMage-MS space allocation strategy can ensure service quality streaming media server at the same time, and can make good use of the disk I/O bandwidth.A new I/O scheduling strategy and algorithm have been proposed, which optimized for VBR streaming media, named VFC. The strategy proposed by the previous FARIMA model and access frequency of various streaming media server file. With use of I/O consolidation, based on FARIMA model to predict traffic of the media file server, combined with FCFS and C-SCAN scheduling ideas, define the VFC, if a media file flow of relatively large, it will give high priority to this request. In the simulation testing, with comparing to the commonly used FCFS and C-SCAN scheduling algorithm, especially for VBR-encoded video file I/O scheduling, the system average response time has improved significantly.Through the above three aspects, we implements the traffic model of VBR multimedia server, a new strategy for scheduling based on traffic proposed. The performance of streaming media server increased significantly.
|
PDF Full Text Request