Research On Large Scale Particle Transport Simulation And Visualization System Based On GPU Clusters

Nowadays,with the rapid development of scientific computing,particle simulation,as one of the most important role,which has an increasing large scale,brings higher demand to computer assisted modeling,scientific computing,and visualization.While the storage and computing resource required by the particle simulating system surpasses single computer's computing ability,High Performance Computing(HPC)and parallel computing makes visualization of massive data simulation possible.Theory and implementation about parallel rendering brings series of classic systems,including the GPU clusters with powerful graphics functions,floating point calculation ability and highly parallel hardware structure,which provide strong support for massive data scientific computing.This thesis is derived from the Chinese strategy project of the Chinese Academic of Science--" Accelerator Driven Sub-critical(ADS)System ".The project aims at using GPU clusters to accelerate the particle translation simulation in the spallation target.This thesis has designed and implemented a set of large-scale particles simulation and visualization system.The main work of this thesisincludes:1.Accelerating tens of millions of particles transport in the spallation target using GPU-CUDA and in situ rendering;2.Proposing a mixed sort-first and sort-last parallel rendering structure based on GPU cluster,which is suitable for GPU clusters and the spallation target simulation;3.Deploying several parallel image composition algorithms,which could achieve good performance.The system is implemented by resource manager torque to assign tasks,then use GPU-CUDA technology to accelerate particle transport.Next realizes rapid rendering on each node through binding particle date on GPUs and the OpenGL buffers,finally employs graphics library OpenCV to implement pixel splicing and depth composing.The system aids researching in particle flow transport simulation in ADS spallation target by means of computers.It combines visualization code and calculate program,directly shares memory data,avoids I/O bottleneck,and gains very high speed-up ratio.