A Prediction-Based Transcoding System for Video Conference in Cloud Computing

We design a transcoding system that can provide dynamic transcoding services for various types of video conferencing clients. It can be deployed to many kinds of cloud computing platform and the scheduler can change the setting of any transcoding task to achieve adaptive transcoding requirement form mobile clients. In this paper, we first analyzed the difference between video conference transcoding and on-demand video. We determined that transcoding service for video conference must be real-time and existing scheduling algorithms are not suitable for video conference transcoding. Therefore, we propose a prediction model that can predict how much computing resources that a transcoding task may require. This prediction model is based on the parameters from source video or the transcoding task. Based on the information from resource monitor and prediction model, the scheduling algorithm can make a more appropriate scheduling decision and keep the load of each virtual machine in transcoding system in a good balance.