Testing weak scalability of parallel solver: cg + hypre_amg

dokken · August 4, 2020, 10:39am

Hypre AMG should not be treated as a black box, and has various options that can be tweaked. As the iteration count is non-constant, you have your first issue.
You should also note that multigrid solvers should be supplied a near nullspace. See for instance:

github.com

jorgensd/dolfinx_mpc/blob/master/python/benchmarks/bench_periodic.py#L214-L228


with dolfinx.common.Timer("MPC: Prepare solver"):
    # Create nullspace
    nullspace = PETSc.NullSpace().create(constant=True)
    PETSc.Mat.setNearNullSpace(A, nullspace)

    # Set PETSc solver options
    opts = PETSc.Options()
    if boomeramg:
        opts["ksp_type"] = "cg"
        opts["ksp_rtol"] = 1.0e-5
        opts["pc_type"] = "hypre"
        opts['pc_hypre_type'] = 'boomeramg'
        opts["pc_hypre_boomeramg_max_iter"] = 1
        opts["pc_hypre_boomeramg_cycle_type"] = "v"
        # opts["pc_hypre_boomeramg_print_statistics"] = 1

for a setup that has a constant number of iterations with hypre-amg.

Topic		Replies	Views
Good choice of solver parameters on parallel Stokes mixed formulation resolution? Linear Algebra	3	781	March 8, 2021
Why does PETSc krylov solver fail on a more resolved mesh on the same exact problem? Linear Algebra	1	809	July 30, 2020
Application of GAMG preconditioner for singular block Linear Algebra	3	826	February 1, 2022
AMG for elasticity problems leaks with PETSc AMG smoothers Linear Algebra	3	278	April 12, 2024
Mumps in parallel slower than in serial	1	3359	April 24, 2019

Testing weak scalability of parallel solver: cg + hypre_amg

Related topics