Research On Real Time Dynamic Scheduling Algorithm In Cloud Rendering Environment

With the rapid development of 3D animation, film special effects, building models, digital media and other industries, people’s demand for three-dimensional scene rendering increase. Three-dimensional rendering has large computation and much more data, especially with the development of modeling technology and peoples’ improving requirements of rendering quality, object models become more sophisticated and the number of geometric models also increasing. The improving rendering demand for light treatment and particle effects which are in the scene makes the rendering a sharp increase in the amount of data and computation.Traditional render farm made a partial solution to the three-dimensional rendering which has large computation, lots of data and cost much time, however, the traditional cluster-based render farm base on common clusters, its scalability poor, remote control inconvenient and very difficulty in data sharing.In this paper, based on the problems in the rendering process we proposed the cloud rendering. Design and implement a set of cloud rendering framework, and make research for the rendering system task scheduling problem, we propose a real-time dynamic scheduling algorithm and finally did some experiments to verify the efficiency of the algorithm. The main work is as follows:1)We proposed a rendering framework based on cloud computing. Through the study of parallel rendering, according to the characteristics of rendering lines we divide the render job into sub-tasks with the sort classification, then do some synthesis after the sub-tasks are completed. In addition, we use cloud computing technology to divide the scene data according to sort and make a distributed cloud storage; dispatch the sub-task in real-time through the message-based mechanism. On rendering nodes, we using container-based technology to encapsulate the rendering task and control the node through REST technology.2)Design a real-time dynamic performance rendering nodes each drawing environment assessment mechanism for the cloud rendering, then on this base to make task scheduling and dispatching. For the case of the network and hardware have uncertain performance in the cloud computing environment, we predict the performance evaluation parameters for each machine based on the historical data and make a comprehensive evaluation for the machine though performance, network status, the data read speed and other parameters. Then proposed a scheduling algorithm for the real-time rendering based on the prediction parameters is, the scheduling algorithm considers the real-time requirements of rendering task by scheduling the sub rendering tasks.