计算科学
巨量平行
计算
并行算法
超级计算机
并行处理
加速
作者
Yu-Yong Jiao,Qiang Zhao,Long Wang,Gang-Hai Huang,Fei Tan
标识
DOI:10.1016/j.compgeo.2018.11.004
摘要
Abstract Large-scale numerical simulation using spherical discontinuous deformation analysis (SDDA) is time-consuming, and this obstructs the application of the SDDA method for simulating the large-scale discrete spherical system. A key factor that affects the computational efficiency of the SDDA program is the solver of linear equations. To solve this problem, this paper replaces the original equilibrium equations solvers in SDDA by the conjugate gradient iterative method which is applicable for parallel computing, and three parallel computing models, i.e. pure MPI (Massage Passing Interface), pure OpenMP and hybrid MPI+OpenMP, are proposed to investigate which is better for SDDA. The parallelization algorithms are incorporated into the original SDDA source code, and the modified program is implemented on the high-performance computing clusters of Sugon TC4600. Three numerical examples are simulated for verification, and the results demonstrate that the proposed hybrid parallelization model is correct and effective.
科研通智能强力驱动
Strongly Powered by AbleSci AI