Parallel Visualization Based on Heterogeneous Computation Resources

Abstract

Visualization and interacting demonstration of vast data is important in many areas. As the data is becoming more and more enormous and complex, it is necessary to employ parallel solution with multi-core CPUs and GPUs to accelerate the visualization and pre-process. This paper focus on parallelizing the visualization of vast data and relevant process on computers with heterogeneous computation resources. A oct-tree and LOD based structure is employed to reduce computation during visualization and interaction, and parallelized algorithms of organizing data based on GPU and multi-threads are designed to accelerate the pre-processing. A production line based algorithm is introduced to make full use of I/O interface to reduce the waiting time for loading data in interaction. All algorithms are designed under the frame of general layered heterogeneous model, making sure of the ease for use and portability. Test with real data proves the parallelized visualization algorithm improves the efficiency of pre-processing and demonstration.