Optimizing Mechanism Of Mass Video Transcoding System Based On Mapreduce

With the wide spreading of intelligent mobile terminals，mobile applications arebecoming more and more important. As a main kind of entertainment，video service playsan indispensable role in mobile applications. Because of the diversities of the mobileterminals and video codecs, the demands for video are different with different terminals,and the demands are different even with the same terminals at different times due to thevariability of mobile network. So the videos should be transcoded to many files withdifferent resolutions and different bitrates, then to provide different quality levels of videosaccording to network status. That is to say, massive video transcoding is needed. However,video files have large amount of data and the process of video transcoding istime-consuming. So the computing resource and network bandwidth should be allocatedrationally to reduce the transcoding time.The data management and task scheduling for mass video transcoding (MVTrans)based on MapReduce applies the Hadoop parallel framework to accelerate the transcodingspeed. It uses the method combined with the data management and task scheduling which isaware of data locality to reduce data migration in the cluster. Firstly, the transcodingprocess is pipelined to make computing resources and network bandwidth be allocatedreasonably. And it distributes the video files to transcoding nodes at video access. To reducethe migration of video splits, the task scheduler reserves the tasks to the transcoding nodesthat the video splits are on. And it manages metadata of video splits, make them split, copyand sort in the cluster according to task scheduling, and splits are copied when it is in thetranscoding phrase. To achieve the full utilization of system resources, the videos are cut upinto different file sizes. Finally, it uses the method combined with genetic algorithm andparticle swarm optimization algorithm to analyze the transcoding logs and optimizes theconfigurations of Hadoop.The MVTrans can shorten the transcoding time as well as raise the throughput capacity.The experiment results show that it can reduce32%less time compared to the currentlyexisting transcoding systems based on MapReduce.