Research On Retained Mode Based Parallel Graphics Rendering

With the improvement of theories and technologies of computer graphics, the technology of graphics rendering is more and more deeply used in the fields of design, entertainment and science and has become an important part of human life. Applications from different fields keep challenging the capability of rendering systems on the aspects of rendering speed, realistics, display size and resolution. Parallel rendering became an important technology to implement high performance graphics rendering by introducing theories and methods of parallel computing into the field of graphics processing.Research on system architecture is the core of parallel rendering technology. Traditionally parallel rendering systems are classified as sort-first, sort-middle and sort-last based on the way multiple rendering pipelines are organized. Plenty of research works have been done under this classification. Nowadays most cluster parallel rendering systems are sort-first ones. The architectures of sort-middle and sort-last are often implemented in hardware graphics systems.However, rendering systems can be classified as immediate-mode and retained-mode by its API's style. We present a new way to classify parallel rendering system: it can be classified as immediate-mode parallel rendering systems and retained-mode parallel rendering systems. The basic difference between the two modes is the strategy of data storing which brings out a serious of differences in the aspects of working process, belonging judgement method and system compality. Plenty of research works has been done to the immediate-mode architecture. This dissertation is devoted to the research of retained-mode architecture. The main contributions are listed as following:First, retained-mode parallel rendering architecture is clearly presented. The potential advantages of retained-mode parallel rendering are pointed out by analyzing and comparing to the immediate-mode systems. Targets of research to retained-mode parallel rendering are then given.Next, some key technologies for implementing a retained-mode parallel rendering system are researched in a deep and systematic way. These technologies are listed as following:1. An object distribution method fit to retained-mode parallel rendering system is presented and implemented which includes the mechanisms of object definition, object creation, remote calling, object deletion and synchronization. For a real retained-mode parallel rendering system such an effective and transparent object distribution scheme is a must.2. There are two different data storing strategies for retained-mode parallelrendering systems: redundant storing and unified storing. A working architecture for unified storing system is presented with a prototype system implemented.3. Research shows that retained-mode systems can do belonging judgement more effectively by using its capability of handling geometry data to divide modelsinto small units. Ball-dividing provides faster belonging computation comparing to common box-dividing. Unified data storing and model dividing are important characteristics of retained-mode parallel rendering systems showing the advantage of retained-mode parallel rendering architecture.4. By analyzing the disadvantage of the parallel rendering load balancing methods based on geometry data analysis, a load balancing method based on time-space conversion is presented which uses time value as the measure of a rendering node's work load and uses space value to control the distribution of work load. A reasonable algorithm is used to convert time value to space value. The main advantage of this method is that its overhead is pretty low. Practical results show that this method is very effective. At last, by using part of the results mentioned before a retained-modemultiple-screen parallel rendering system called MSPR is implemented which candeliver large screen and high resolution graphics rendering.