Research Of Parallel Streamline Tracking Method And The Simulation In Tóthian Basin Groundwater

Reservoir engineering and groundwater research often use the particle tracking method to achieve the purpose of streamline simulation,and use the distribution of streamlines to reflect the shape of the real flow field.Thus,it provides researchers with intuitive flow information.As the research area becomes larger,the number of streamlines increases rapidly.When using a common processor to calculate millions of streamlines,the calculation speed is very slow,and the calculation time is so long,which brings great inconvenience to large-scale engineering applications and groundwater research.In order to meet the real-time demand of streamline generation that practical applications request,this paper proposes to accelerate the simulation of streamlines.In this paper,the streamline simulation based on particle tracking method is analyzed.It is found that the calculation between streamlines is completely independent,and there is no sequence and no data dependency.Therefore,this paper proposes to implement streamline in parallel by using parallel techniques.This paper mainly uses three parallel mechanisms which include MPI,OpenMP and CUDA to realize streamline simulation in parallel based on single-node and multi-node.The paper takes three-dimensional Tóthian basin as an example,to analyze the performance of the parallel program.On a single-node computer,the multi-core of CPU,single GPU,and multi-GPU,used as hardware accelerators,are respectively responsible for parallel computing of streamlines.Experiments show that the acceleration effect of parallel program based on multi-core technology is general,and its speedup is limited by the number of CPU cores.The programs based on GPU can quickly achieve better speedups.When calculating 100 million streamlines,the speedup of a single GPU is approximately 160 times,and six pieces of GPUs reach 630 times.Experimental results prove that,for 1 to 6 GPUs,the more GPUs are used,the better acceleration effect we can obtain.In multi-node cluster system,multi-processor and multi-GPU,used as computing resources,respectively perform parallel computing of streamlines.In the program which is based on multi-processor and multi-core,several processes are distributed on multiple nodes,and each process has 40 threads.When using 32 processes,which are distributed on 16 nodes,the speedup reaches 170 times.The experimental results show that the parallel program has good scalability and is suitable for porting to large-scale cluster systems.The parallel program based on multi-processor and multi-GPU is completed by MPI,OpenMP and CUDA.In the programming,each process uses four threads,and each node has two processes,to make full use of 8 pieces of GPU on each node in AI service platform.The program obtains the best acceleration effect when using 4 nodes,8 processes,32 threads and 32 GPUs.The whole experimental results show that the parallel program based on GPU has poor scalability.