| With the widespread use of mobile phones, tablets and other intelligent mobile equipment, shooting video has become very easy. Meanwhile, the wireless network coverage is increasingly wide and the number of video service platform is larger. Watching and sharing video have become important daily entertainment. A lot of video data are generated every day with the widely used of monitor equipment. However, these video files usually need to be transcoded to meet the needs of transmission in different environments or the demands of playing videos in different terminal equipment. The traditional video transcoding is centralized. Thus, its storage capacity and computing power are very limited and cannot be optimized with the increasing data dynamically. Therefore, it’s not suitable to deal with large-scale video data. This thesis proposes a distributed transcoding system to support a large-scale video transcoding.The thesis first presents a detailed introduction of the two core components of Hadoop cloud platform, i.e. distributed file system HDFS and distributed programming framework MapReduce. This thesis is focused on introducing the design of HDFS and its reading/writing procedure, the implementation principle and operation process of MapReduce. Then several commonly used Hadoop schedulers are introduced. Besides, this thesis introduces some relevant knowledge on the video compression and transcoding. On the basis of the understanding of video transcoding and Hadoop cloud platform. A distributed transcoding system is designed to support the transcoding of large-scale videos. The system contains four modules: service module, management module, transcoding module and storage module. The design and implementation of each module are described in detail. Finally, this thesis implements the proposed system through using: 1) the distributed storage capacity of HDFS; 2) the parallel processing ability of MapReduce; 3) the video processing function of the multimedia library Xuggler. The system is mainly composed of Hadoop cluster which can be built with the ordinary PC. The cluster’s ability of storing and calculating can be improved by increasing the node amount, which makes the system can dynamically adapt to the transcoding of the growing large-scale video data.The last part of this thesis first makes a preliminary functional test to the system and several groups of different sized data sets are used to test the transcoding performance of the distributed transcoding system, especially the total transcoding time of the system. The test results show that the system has good parallelism while transcoding large scale of videos. Furthermore, the overall transcoding time of the system can be greatly reduced with the expansion of Hadoop cluster. The thesis also analyzes several configuration options of HDFS on affecting the performance of the transcoding system. Experimental results indicate that when HDFS block size is close to the size of the most original videos and when the replication is 3, the system performs better in terms of total transcoding time. |
