相场模拟的并行计算方法.pdfVIP

Hybrid MPI + OpenMP Approach to Improve the Scalability of a Phase-Field-Crystal Code Reuben D. Budiardja reubendb@ ECSS Symposium March 19th, 2013 Project Background 飩 Project Team (University of Michigan): Katsuyo Thornton (P.I.), Victor Chan 飩 Phase-field-crystal (PFC) formulation to study dynamics of various metal systems 飩稯riginal in-house code written in C++ 飩 Has been run in 2D and 3D systems 飩稴olves multiple Helmholtz equations, a reduction, then an explicit time step 2 Solving the Helmholtz Equations 饊嫽2鈭 + 饊倢饊彁 鈭 = 饊弾 飩稯riginally used GMRES with Algebraic Multigrid (AMG) preconditioner from HYPRE 飩 In 3D, discretization matrix is large and may become indefinite 飪 difficult to solve, requiring large iterations 飩 Poor weak-scaling results 飩 Prohibitively long for indefinite matrix case 飩 Increasing memory requirements with iteration 3 Goal 飩稴calable to solve larger problem 鈥 Weak scaling: maintain the time-to-solution with increasing number of processes and a fixed problem size per process 飩 Decrease the time to solution to 1 sec / time step 鈥 Strong scaling: decrease time-to-solution with increasing number of process and a fixed problem size 鈥 Exploit other parallelism (with OpenMP?) 鈥 Investigate better preconditioner 鈥 Different method (library?) to solve the equations 4 Goal 飩稴calable to solve larger problem 鈥 Weak scaling: maintain the time-to-solution with increasing number of processes and a fixed problem size per process 飩 Decrease the time to solution to 1 sec / time step 鈥 Strong scaling: decrease time-to-solution with increasing number of process and a fixed problem size 鈥 Exploit other parallelism (with OpenMP?) 鈥 Investigate better preconditioner 鈥 Different m