Architecture for Predicting Live Video Transcoding Performance on Docker Containers

Video can be streamed live with differentapplications (e.g. YouTube Live, Periscope). Typically, thevideo content is adapted for end users based on receivingclient’s capabilities, and network bandwidth. The adaptation isrealized with different video representations, which are createdby transcoding the original video content. When video isstreamed live, transcoding has to be completed within real timeconstraints, which is a computationally demanding process.Particularly, live transcoding should be enabled efficiently by acontent distributor to minimize resource provisioning costs.The contribution of this paper is an architecture for predictinglive video transcoding performance on a Docker-basedplatform. Particularly, cloud resource management for livevideo transcoding has been focused on. A model was trainedbased on measurements in different transcodingconfigurations. Offline evaluation results indicate that livetranscoding speed or CPU usage can be predicted with 3-8 %accuracy. When video is transcoded on virtual machines basedon predictions in a prototype system (live), live transcodingspeed prediction accuracy is within a similar range as theoffline performance, but worse for CPU usage prediction (5-15%). In most cases the specified range for transcoding speedand CPU usage can be achieved at least with a precision of 76%