Containerized Media Cloud And Its Resource Scheduling Problem For Distributed Media Transcoding

Video is the most important data type in daily life,providing information in a very intuitive way.In the era when information is developing rapidly,media data is changing its pattern of appearance.However,productivity of video is becoming deficient and has become a bottleneck of the media supply chain.Video production over cloud is a good solution to the lack of productivity in local and cluster environments.Cloud computing can offer massive computation,storage and network resources,and support pay by usage mode to best fit for the requirements like elastic resources and low cost in media production scenario.Besides,the popular virtualization technology,container,can offer far higher hardware resource utilization and more flexible system deployment than traditional virtual machines.And MicroServices,as a hot architecture for native cloud application,can provide faster upgrade,agiler maintenance and better scalability to cloud applications.To solve the lack of productivity on high quality and new type of media contents,this paper extends and upgrades existing distributed media transcoding systems.By using container technology and MicroServices architecture,a containerized media transcoding cloud is designed.Task executing process and message transmission mechanism are redefined to achieve loose coupled modules packed up by Docker.Comparing with traditional system,it is lightweight,easy to extend,agile to maintain and has higher efficiency.We deploy the system on TianHe-II Super Computer Cloud Platform and test the system with fixed GOP frame rate up convert usecase in IPTV to show its efficiency and correctness.Service discovering problem is unavoidable in MicroServices systems and can increase communication and network consumption when scheduling resources and tasks in container media transcoding cloud,thus limit scalability.Aiming at this problem,this paper defines the concept “cluster” as a group of symmetric transcoding instances to transparently and self-orgainzed providing service,according to the system structure.Then a service discovering model is designed to discover and maintain the services in a layered and decentralized manner.The model can reduce additional resource consumption introduced by service discovering.We prove the properities of the model by graph theory methods and analyze it with large scale short video fast transcoding scenario in social networks.