Research On Parallel Volume Visualization Technology For Large Scale Scalar Filed Dataset Based On OpenMP/MPI

Volume visualization is a significant 3D scalar field data visualization method,which directly generates semi-transparent 2D images from volumetric data.However,the existing serial volume rendering algorithms are inefficient and cannot meet the large-scale scalar field data volume visualization needs.In this paper,we proposed parallel volume rendering algorithm with thread/progress parallel method to address large-scale scalar field data volume visualization.The main contributions are as follows:(1)Multi-threaded parallel projected tetrahedral algorithm.Visibility sorting and tetrahedral projection classification/decomposition are the most time-consuming parts.This paper optimized the two parts with Open MP.A 20-core CPU experiments show: compared with serial algorithm,the proposed algorithm can achieve a speedup of 3.5,and outperforms the fastest VTK implementation at a speedup of 2.5.(2)Multi-progress parallel volume rendering algorithm.This paper leveraged MPI framework and implemented multi-progress parallel volume rendering algorithm.On the basis of parallel loading large-scale volumtric data;each process splits volumtric data with KD tree algorithm;then,each process uses an independent pipeline to calculate volume rendering images;and final result was synchronized with a tree composite strategy.A 80-core CPU experiments show: compared with serial algorithm,the proposed algorithm can achieve a speedup of 37.6.(3)Loosely-coupled C/S based parallel volume visualization system.Client generates interactive commands and displays volume rendering results,and server handles visual computing and rendering tasks.Client/server communicate through socket;and MPI is leveraged communication among server processes.Design pattern application and modular implementation ensure high flexibility and scalability of the system.Inter-node and cross-node experiments show: the proposed system can be well applied to large-scale unstructured grid data volume visualization,and meet real-time interaction needs.