Study Of Hadoop-based Video Transcoding Optimization

Under circumstances of “Internet Plus”, people have the growing demand for video transcoding which with higher speed and quality than before. Based on the report of 2015 Annual Summary for Network Traffic, video created the largest proportion of the total network traffic consumption. For now, both industry and academic world are concentrating on how to transcode video efficiently with high quality and high-availability. This report focused on the research for optimizing video transcoding process.To obtain an efficient process and an accurate result, the experiment used cloud computing method to transcode video. This method could enable the cloud to do the parallel computing. The report also designed a cloud platform of video transcoding. This platform used a three-level structure: Iaa S, Paas and Saa S. We chose Amazon for Iaa S level, Hadoop for Paa S level and a high performance video transcoding application for Saa S level. The report also introduced a video transcoding optimization strategy, which combined the FMPEG and the Map Reduce technic.This report mentioned an architecture called S_Map Reduce. Introducing the virtual IP mechanism into S_Map Reduce could improve many performances such as increasing scalability and making the Map Reduce architecture able to cross-domain. A heart beat process was also designed in the S_Map Reduce architecture, which ameliorated the load –balance module and made the whole architecture more stable than before. This process enhance the computing architecture performance by optimizing the Map Reduce’s work-flow and data-flow process.The video transcoding is based on the cloud platform, which combines the FFMPEG and S_Map Reduce to implement the distributed transcoding. The main process include :(1) According to the fixed scene and several limited conditions, dragged part would be deleted to weigh properly for FFMPEG. The design of NALU has been completed and anti-competition mechanism has been put forward to avoid competition and displacement error.(2)APU format has been designed to match Map Reduce to deal with in block, optimizing the decoding process and improving the efficiency(3)Re-check algorithm has been designed to ensure the integrity and order of the data.The whole experimental environment used 19 spot to build the transcoding cloud. The data in this experiment was chosen by multi-level and multi-dimension to insure that this test was universal and practical. Experiments were testing on the video transcoding performance and the after transcoding video quality. The result showed that under the cloud platform of video transcoding, the video transcoding became more efficient with the data size increasing. By using multi-spot method, when the single file size is 10 GB, the efficiency promote to 16.4 times than the single-spot method.