Parallelization of groundwater flow simulation on multiple GPUs

GPU has been applied in groundwater flow simulation. In order to improve the performance of the GPU groundwater simulation further, this paper studied the method of parallelizing three-dimensional groundwater flow simulation on multiple GPUs. The most time-consuming part in the groundwater flow simulation, solving equations, is parallelized on multiple GPUs. The PCG solver is parallelized to solve equations. To maximize the communication efficiency of the parallelized PCG, optimizations that reduce the unnecessary data communication and overlap the communication with calculation were used. Experimental tests using 6 NVIDIA K40m GPUs show that the maximum speedup of the PCG solver and the groundwater flow simulation is up to 36.3 and 11 respectively for a steady-state simulation.

[1]  Xiaohui Ji,et al.  A simulation of large-scale groundwater flow on CUDA-enabled GPUs , 2010, SAC '10.

[2]  Yousef Saad,et al.  Iterative methods for sparse linear systems , 2003 .

[3]  Hiroshi Okuda,et al.  Conjugate gradients on multiple GPUs , 2010 .

[4]  Zhiqiang Xie,et al.  A Jacobi_PCG solver for sparse linear systems on multi-GPU cluster , 2016, The Journal of Supercomputing.

[5]  Xiaohui Ji,et al.  CUDA-based solver for large-scale groundwater flow simulation , 2011, Engineering with Computers.

[6]  Sebastian Schops,et al.  Multi-GPU Acceleration of Algebraic Multi-Grid Preconditioners for Elliptic Field Problems , 2015, IEEE Transactions on Magnetics.

[7]  Guomin Li,et al.  A Parallel PCG Solver for MODFLOW , 2009, Ground water.

[8]  Junqi Huang,et al.  An assembly model for simulation of large-scale ground water flow and transport. , 2008, Ground water.

[9]  Arlen W. Harbaugh,et al.  A modular three-dimensional finite-difference ground-water flow model , 1984 .

[10]  Xiaohui Ji,et al.  Parallelization of MODFLOW Using a GPU Library , 2014, Ground water.

[11]  Jingli Shao,et al.  Accelerating Groundwater Flow Simulation in MODFLOW Using JASMIN‐Based Parallel Computing , 2014, Ground water.

[12]  Jiaquan Gao,et al.  A multi-GPU parallel optimization model for the preconditioned conjugate gradient algorithm , 2017, Parallel Comput..

[13]  Arlen W. Harbaugh,et al.  MODFLOW-2000, The U.S. Geological Survey Modular Ground-Water Model - User Guide to Modularization Concepts and the Ground-Water Flow Process , 2000 .

[14]  Scott B. Baden,et al.  Effective multi-GPU communication using multiple CUDA streams and threads , 2014, 2014 20th IEEE International Conference on Parallel and Distributed Systems (ICPADS).

[15]  Arlen W. Harbaugh,et al.  A generalized finite-difference formulation for the U.S. Geological Survey modular three-dimensional finite-difference ground-water flow model , 1992 .

[16]  Jeremy T White,et al.  Use of General Purpose Graphics Processing Units with MODFLOW , 2013, Ground water.