Mobile Streaming Oriented Cache Prefetching Strategy

With the continuous development of wireless networks and the performance of smart mobile,mobile Internet and mobile applications have got widespread concern in the Internet community.As the "Killer Application" of traditional Internet,streaming service has also been popularized in industry and that makes it one of the mainstream mobile Internet services.Traditional mobile streaming media services are typically based on the system ar-chitecture model of B2D(base-station-to-device).howerver,because of the limit of the base station bandwidth's bottleneck,it lacks system scalability.In recent years,with the popularity of WiFi in smartphones,D2D(device-to-device)model are widely known as an important way to enhance the throughput and quality of mobile service.Based on this,our project team designed and implemented a system called "Smartphone-Oriented Collaborative Mobile Streaming System" which supports smartphones that are embedded with WiFi to construct MANET and cooperative data transmission and adaptive data transmission with WiFi that can effectively alleviate the bandwidth of cellular network base station.Caching strategy is very important in "Smartphone-Oriented Collaborative Mo-bile Streaming Service System".It is very necessary to design effective caching place-ment and prefetching strategy.The article focuses on the prefetching strategy based on collaborative technology of D2D and can be divided to two parts.1)The prefetching strategy based on overhearing system.When playing a single video(usually long video),users ofen don't complete playing in order.Sometimes they do some VCR operations such as forward,backward,replay,etc.Those are user behaviors which we will consider in this paper.According to the playing history of a user,we model their behavior to predict the probabilities to play in next time slot of different data blocks.In this paper,we combines the specific overhearing-based mobile streaming system and user behavior model based on reinforcement learning to test and analyze the performance.The results showed that prefetching strategy based on overhearing system can help to reduce the communications between nodes and server and increase data transmission efficiency.2)Matrix factorization based prefetching mechanism based on time and space.In a mobile streaming scene,multiple users often watch different short videos.The article combines the properties of time and space in wireless MANET environment with matrix factorization based prefetching mechanism to design the algorithm to op-timizatize the efficiency of WiFi.The results showed that when considering time and space properties,we can make use of the WiFi communication capabilities to improved interoperability of the nodes,meanwhile,the new algorithm can meet the accuracy re-quirement of the basic top-N prediction algorithms.