Research On Storage Of Streaming Media In Cloud Platform

On video streaming system, video data caching strategy affects the quality of the userexperience. Due to the heat of each video are different, the video which has high heat needed togenerate a replica of the video and the video which has low heat needed to will have to be deletedand replaced. Currently there two kinds of cache: part of the cache and cache completely. For partof the cache, although the large number of video could be cached, initial moment on demand ratio ishigh, but the rest part of video needed to be send from the source server, which is easy to achievethe source server bottleneck of I/O throughput. For fully cache, a limited number of video can bestored. This paper combined the features of two cache ways, researches mixed combination strategyof part and completely cache and related technologies to improve the performance of the system.These innovations are briefly summarized as follows:(1) The generation of replica: firstly we put forward a kind of generation and cachingalgorithm, which determine the parts of video prefix dynamically. The simulation results show thatcompared with the fixed prefix cache algorithm, the dynamic prefix cache algorithm effectivelyimprove the hit ratio of video on demand. Secondly We present a prediction algorithm which isbased on the Last-k time windows, and then the number of replica of each video required iscalculated more accurately.(2) The replacement of replica: firstly we present an optimal server selection algorithm basedon weighted entropy. Compared to other algorithms such as the round robin, least load first, leastconnections first, the proposed algorithm considering three factors of the balance of storage,balance of heat, and balance of bandwidth, implements the comprehensive optimal systemperformance. When the space is insufficient, a smooth deletion strategy is presented to achieveoptimal system stability, which considering the generation time a copy of the video, a copy ofcapacity, changes in heat fluctuations. Simulation data shows that the algorithm can effectivelyavoid the fluctuations of reject rate due to accidental deleted replica and ensure the QoE.(3) On the basis of the above research results, we put forward mixed combination strategy ofpart and completely cache. We do experiments of the generation, the placement and the cache ofvideo by the hybrid strategy. Simulation results show that the proposed strategies and methods improve the performance of the system, and improve the QoE.