Distributed Real-Time Transcoding For Mobile Videos

With the rapid development of mobile Internet, for users, watching online videos any-time and anywhere has become an urgent requirement. However, the limited resources onmobile devices such as mall screen sizes, embedded processors and limited battery capaci-ties make them hard to play Internet videos like desktop and laptop PCs. Hence video tran-scoding to specific devices is needed. Most existing video transcoding systems are usingoff-line transcod ing which only transcoding video into a fixed format and resolution, andsome real-time transcod ing strategy have not yet considered the video playback on theclient. How to provide real-time transcoding service while the client is playing the tran-scoded video and effectively solve the low flexibility, storage pressure, and poor user expe-rience of existing transcoding system has important theoretical and practical significance.According to experiment we find that the playback time is greater than the transcodingtime and the network transmission time when transcoding video from high-quality tolow-quality for mobile devices. Thus, it is feasible for real-time transcoding in distributedcomputing environment. The key lies splitting video dynamically based on network condi-tions. In this paper we propose and implement the system CloudTranscoder which based ondynamic slice cutting algorithm to reduce the start-up time and increase the flexibility ofreal-time transcod ing. Then we constructed a task scheduling model which take the time oftranscoding tasks into concern, the model takes full advantage of the cloud computing re-sources on each server and ensure the tasks are in real-time. In order to ensure the videoclips which are transcoded in different nodes generate consistent video rate, we improvedTM5rate control algorithm. We use the HTTP protocol to send the video data which hasfinished transcoding to users in order to provide users with streaming services, any clientsupport HTTP streaming can playback the transcoded video in our system.Experimental results show that the response time for users to watch the transcodingvideo is less than5seconds and interruption rarely occurs in CloudTranscoder. After usingdynamic slice cutting strategy, the fragment size change with client network conditions,produceing strong flexibility. The system provide personalized real-time transcoding ser-vice for different smartphones with only one copy of video source stored on the server, andthe average load difference between transcoding servers is about10%.