Based On The Sort First Parallel Rendering System For Dynamic Load Balancing

With the continuous upgrading of graphics hardware and the increasing development of computer graphics technology, computer graphics applications have been more and more closely related with people's daily lives and scientific researches. Besides the Traditional graphics applications, the field of computer graphics applications has been extended to space exploration, the military battlefield simulation, film special effects, entertainment, games and medical imaging. New technology brings about people's demands to the high quality graphics applications services. The expensive specialized graphic workstation can not meet more and more people on the high-definition, real-time interactive and realistic graphic application needs.Interactive graphics applications have long been challenging graphics system designers by demanding machines that can provide ever increasing polygon rendering performance. Another trend in interactive graphics is the growing use of display devices with pixel counts well beyond what is usually considered"high resolution". Among the parallel rendering architectures, we discover the Sort-first is the only one architecture that deliver millions of rendered pixels for thousands of primitives in real-time without requiring phenomenal communication bandwidths and excessive duplication of hardware. In addition, it features reasonable scaling characteristics. Therefore, Sort-first has become the preferred choice in parallel rendering applications.Because of the advantage of the frame coherence and the implementation without intervening during the transformation phase and rasterization phase in the pipeline, Sort-first may reduce the extra unnecessary workload with low communication requirements. However, it is susceptible to load imbalance as primitives may clump into regions to concentration the work on a few rendering node. To solve this problem, many algorithms have been proposed, but these algorithms can be fully functioning on the premise that the performance of each rendering node within the system exactly the same. All rendering nods must be set up with the same hardware and software configurations so that they have the same performance. When not in use, this will result in waste of resources. In fact because the state of each node is not necessarily exactly the same, so the above homogeneous node-based load balancing algorithm for parallel rendering is not practical.To address the problem above, the adaptive load balancing algorithms for Sort-first parallel rendering have been in-depth analysised and studied based on the projects in this paper. The main tasks are as follows:1. Analysises and studies the adaptive load balancing algorithms for homogeneous-node parallel rendering algorithm based on the last load feedback. The algorithm is applied into the proposed system in order to get the performance increase.2. Proposes and implements the load balancing algorithm for parallel rendering based on heterogeneous-node system. The algorithm fully takes the performance of each node in the system into account so that each node are assigned the rendering task according to its capacity, improving the efficiency of the whole system, decreasing the probabilities of the load imbalance. And the algorithm is successfully applied to the system.