Key Techniques Study Of Parallel Splatting Algorithm On Cluster

Volume visualization especially volume rendering has been the most important and rapid developed techniques in the visualization of three-dimensional data sets recent years. Object-order volume rendering (Splatting) method is a volume rendering technique for reconstruction in feed-forward volume rendering. It has the advantages including simple voxel traverse, easy use of all kinds of speedup techniques, high parallel ability, and easy hardware implementation. It is very suitable for the large-scale data parallel visualization computation and can be used in the design of accelerative hardware. How to bring the parallel computing techniques into Object-order volume rendering (Splatting) method is discussed in this paper。First, based on the theory of parallel Splatting algorithms, the technology of accelerating Splatting,the design pattern and analyses of Splatting algorithms,the parallel computing model of Splatting were studied in this paper.Second,to study furtherly on parallel Splatting algorithm, in this paper based on cluster, the computational mode of the object data parallelism Splatting algorithm was studied. And the parallel strategy that can effectively improve the rendering speed and the algorithm efficiency was designed on cluster. In this strategy, the border effect problem was solved thoroughly. To solve the problem that in the Master-Slave mode, which is used very extensively in Splatting parallel algorithm, the task of Master node becomes too heavy with the increment of the volume data, a dynamic parallel Splatting computational mode based on cluster was designed and studied in this paper to raise the computing speed of Splatting algorithm. It used the Master-Slave-Gleaner computation mode adding task pool and result pool--the strategy of the dynamic load balancing. The results of theoretical analysis and numerical testing showed that the computational mode presented can decrease computing time, increase accelerator, improve computing efficiency greatly, improve the rendering speed effectively without reducing image quality and attain the gain of the load balancing effectively. In addition the MSG model is suitable for a wide variety of volume rendering algorithms and has good scalability.At last, the dynamic parallel MSG computational mode was used to render the large volume data of the laboratory and the result data show the practicability of the mode.