Research On DASH Optimization Strategy For Multi-server

Traditional RTP/RTSP-based streaming media technology has the disadvantages of inability to penetrate firewalls or NAT and satisfy the specific need for server configuration, which greatly limits its application. In recent years, Dynamic Adaptive Streaming over HTTP (DASH) emerged, it can dynamically adjust the video bitrate of user requests and improve the utilization rate of the network depending on the current network condition. Therefore, this technology draws wide range attention in business and academia. With the expansion of scale, DASH service model also began to change from single server to multiple servers. In the multi-server scenario, client can simultaneously download video fragments from multiple servers. Compared with single server, multi-server can offer expanded bandwidth, link diversity and reliability.In the multi-server scenario, the client's video quality would fluctuate much frequently due to the user's high dynamic request and multi-server's different transmission bandwidth. Such fluctuation would affect the stability of the client's video quality and deteriorate the quality of experience(QoE) seriously. The existing solution mainly focus on ameliorating client adaptive algorithm, which treats the fragment as the smallest unit to complete the bit rate adaptive process, it is often difficult to achieve a good result in terms of stability of the video playback quality. Therefore, this paper mainly pays attention to the DASH client quality fluctuation problem in multi-server scenario, and proposes a DASH video transmission framework based on Software Defined Network (SDN). The contribution of this work are mainly three-fold:1) building a DASH system service model based on SDN in multi-server scenario.By employing the openflow controller to collect the available bandwidth of each server,the web proxy server forwards the client request to the openflow controller, which executes the fragmentation scheduling algorithm to assign the service request to the client.2) A video fragment scheduling algorithm with the granularity of video block (one block includes multiple video slices) is proposed. By considering several video segments as one video block, these segments are assigned to a series of servers through an allocation strategy. This two-step method can not only realize the bit rate synchronization of the video segments in the same video block, but also enable the segments to be downloaded in an orderly manner.3) Q learning algorithm is adopted in the client with the environment state of bandwidth, client buffer and video bit rate. A QoE-related reward function if firstly constructed, and then the Q matrix is generated by training and learning. The Q matrix will be the decision basis of the client bit rate.Finally, the validity of the whole system is verified by experiments. The results show that this strategy guarantees the video fragment orderly downloaded depending on the service bandwidth, reduces the frequency of video quality fluctuation, increases the stability of client playback and improves the quality of client video.