The Design And Implementation Of Massive Video Real-time Transcoding System

Nowadays, with mounting number of videos needed by the Internet, broadcast network, telecommunications network and innate requirement of video applications for multi-format multi-platform videos, existing transcoding system has been much too awkward to handle such situation. As the Cloud computing popped up, with tremendous number of fans contributing increasing number of new techniques, processing of massive video has been much easier than before. But existing cloud transcoding systems are usually unsatisfactory in fulfilling users’expectation of transcoding time, for users are often frustrated to find that the transcoding has not been finished after quite a long time. So realtime transcoding system is a better choice.This paper describes the cloud tool kits openstack and hadoop Combined with the description of transcoding system and the gradation framework of cloud system, this paper lay openstack down as the Iaas level of the system, handling the computing and storing task,and has hadoop be the Paas level, handling distributed transcoding business, and use the the mature message queue RabbitMQ structures shared messaging platform for inter-process communication.The introduction of the system designing and implementation is based on the logic level of the system:the control layer, business logic layer, resource layer, as well as four parts:user management, image management, cluster monitoring, task management, as well as independent functional modules the processing module, the video processing module, the message module.In this paper, we use OpenStack cloud tools to build scalable cloud infrastructure management platform, using hadoop,configured with real-time scheduling strategy, to construct real-time cloud transcoding system, using RabbitMQ message queue for inter-process communication, and it’s able to undertake multi-user and massive video. The system use the computing components, nova of openstack to manage resource pool, which is in charge of dynamically creating and destroying virtual machine node, ensuring the resource layer efficient and stable. In the meanwhile, it’s equipped with multi-level fault-tolerant processing and message queue mechanism to ensure the processing flow of the business system correctness and transcoding stability of the system.Finally, we build a experimental OpenStack system on which to build a Hadoop cluster and tested the ability of resource dynamic scalability and Hadoop real-time scheduling of the system to confirm the correctness and efficiency of the system.