Research And Implement Of Energy Saving Control In Rendering Cluster

After some traditional media such as newspaper, magazine, broadcast and television, Internet-delegated digital media are getting more and more popular. For the sake of springing up of cartoons and high requirements of 3D cartoon production, the inefficient and heavy problems of rendering tasks become apparent increasingly. To resolve the difficulties above, cluster rendering technology emerges as the times require, the reasons of which are that cluster computation contains excellent performance and appropriate cost performance. Therefore, cluster rendering technology has been initialized in scientific computation and resource management all over the world. At the same time the vast energy consumption of render clustering becomes one of the focuses of this field. As the resource turns less and less, and the requirements for the humans control of cluster as well as energy saving become higher and higher, it is urgent to reduce the cluster consumption to the maximum extent under the circumstance of the guarantee of cluster loads and throughputs.In this paper, some researches have been done, for instance how to conduct energy saving control effectively to save electricity and prolong the life-span of rendering nodes in 3D Cartoon Cluster Rendering Systems. Firstly, some concepts of this area have been introduced. Also cluster rendering and rendering energy saving control have been reviewed, as well as the development, actuality and future pulse of current cluster rendering soft wares. Alternatively, the architecture of 3D Cartoon Cluster Rendering System-CCRS has been proposed, based on which tri-layer structure, flow of work and the way of realization of CCRS have been analyzed and the function modules of CCRS have been depicted in detail. Pointed at the features of 3D rendering, a rendering cluster energy saving control frame has been put forward, based on the prediction of the number of the tasks and using automatic power-on and power-off control as the means. According to the history data of rendering tasks combined with the analysis of the cost of rendering node's power-on and power-off and the state conversion rules of rendering nodes, multi-step prediction automatic power-on and power-off control algorithm has been brought. Moreover flexible control strategy and corresponding control method have also adopted aimed at different periods. In this thesis, on-line prediction model has been built on a basis of subtractive clustering off-line RBF neural network prediction algorithms. In addition, the thesis has validated the efficiency of the prediction algorithm and the energy saving control, and analyzed the shortcomings.