A Rotate-Tiling Image Compositing Method for Sort-Last Parallel Volume Rendering Systems on Distributed Memory Multicomputers

The binary-swap (BS) and the parallel-pipelined (PP) methods are two well-known image compositing methods for sort-last parallel volume rendering systems. However, these two methods either restrict the number of processors to a power-of-two or require many communication steps to transform image data that results in high data communication overheads. In this paper, we present an efficient image compositing method, the rotate-tiling (RT) method, for sort-last parallel volume rendering systems on distributed memory multicomputers. According to the number of initial blocks of a partial image, the number of processors, the image sizes, and the characteristics of parallel machines, the RT method can fully utilize all available processors and minimize the data communication overheads. To evaluate the performance of the RT method, both theoretical analysis and experimental test of the BS, the PP, and the RT methods are conducted. In the theoretical analysis, we derive the best performance bound of the RT method in terms of the number of initial blocks of a partial image, the number of processors, the image sizes, and the characteristics of parallel machines. In the experimental test, we implemented these three image compositing methods on an IBM SP2 parallel machine and a PC cluster. The experimental results show that the RT method outperforms the BS and the PP methods for all test samples and match the results analyzed in the theoretical analysis.