User Experience Oriented Schedule Mechanism Optimization Research In File Distribution System

Real-time and non-real-time file distribution services such as:live streaming, video on demand, video conference, software downloading and game upgrading have already dominated internet traffic flow. With the rapidly increasing of large-size and high-bitrate HD videos, more and more excellent quality of experience are required by users in real-time and non-real-time file distribution services. How to optimize the file distribution system so as to satisfy user’s demands for quality of experience becomes a hotspot appealing academia and industry.Through measurement and theory analysis on video replay process, user demand, cache and user resource configuration and cloud bandwidth allocation in large scale and real file distribution systems (e.g., PPTV and QQ xuanfeng system), it is found that the cloud bandwidth influence on swarms is ignored in existing cloud bandwidth allocation research, user demand prediction is still unsolved, the pressure of cloud is too high under Flash Crowd, besides, the video replay process is often interrupted under wireless channel. This paper focuses on schemes improving users’ experience and saving system resource from four aspects such as:client experience optimization, cloud resource rental, caching resource utilization and cloud bandwidth allocation. Then the effectiveness of these schemes is verified through theoretical analysis and experimental simulation.In general, the main contents and innovation points are concluded as below:1) User experience Optimization:through measurement in real systems, it is obvious that existing dynamic adaptive streaming over http technology cannot adapt to the frequent changes of wireless downlink speed, therefore, the video replay process would be interrupted. Existing methods rely on historical knowledge of previous results and current downside speed of wireless channel in order to make video quality selection strategy. However, due to frequent changes in conditions of wireless channel, the bitrate switches frequently. In this paper, under the condition that the future changes of channel information can be inferred through a wireless channel model, the optimized bitrate selection is determined to optimize user QoE based on nondeterministic state decision process. Finally, a heuristic algorithm is designed to approximate the performance of the optimized algorithm.2) Personal demand and popularity forecasting:based on the detailed user historic behavior analysis, recommender methods could predict the resources a user wants with high accurancy, but the prediction result could not tell us when a user needs the resources. Therefore, a personal timeliness demand prediction method is proposed, which is dependent on the correlation between different episodes and users’ watching modes. The algorithm cloud predict the episodes a user wants to watch tomorrow by using machine methods. Finally, a popularity predictions method is designed to approximate the real value, and the result is 12% better than that of pure ARIMA algorithm.3) Caching and user resource utilization:a lot of users request for popular files within short time when new files are released, so the cloud suffers from the heavy load. The pre-push strategy which deploys file on the clients in advance could be used to decrease the cloud pressure with the help of P2P capacity. However, traditional methods don’t give considerations to the personal demand but just simply select users by their historic online behaviors and client performance. Through our measurements, it is found that many people would drop the drama after watching one or two episodes, transmitting the file to users who don’t want the resource will waste the cloud resources. Based on the user demand prediction, a scheme is designed to schedule the cache resources and user resources, and a pro-active algorithm is proposed to minimize the cloud load. The simulation result shows that the scheme could save 40% traffic.4) Cloud bandwidth allocation among swarms:P2P capacity is not stable, and is different among swarms, with the help of cloud bandwidth, the Cloud-P2P hybrid system could guarantee user’s experience. Existing cloud bandwidth allocation algorithms don’t take the cloud influences on user’s downloading lifetime and his uploading speed into account. Based on the fluid model, the relationship between cloud bandwidth and user’s downloading speed is discovered, finally a bandwidth allocation algorithm is designed to optimize users’ QoE in the system.