A general parallel process

General
1

Hi

I would like to solve a simple Poisson problem with two distinct parameters. The two resulting problems (for each parameter) should be solved in parallel, with each of these problems being solved again in parallel. To (try to) clarify, assuming I have 4 cores available, 2 cores would be used for each of the problems (which are being solved in parallel).

My question is aligned with a previous post: How to solve in parallel for parametric study, with the difference that I wish for each of the problems to be solved in parallel (domain decomposition).

I’m not sure if it’s possible to do this. I tried using comm.Split , but the results are not correct.
I appreciate any help, as this would greatly assist me. Thank you.


# set MPI communicator
comm = MPI.comm_world
size = comm.Get_size()
rank = comm.Get_rank()
# split comm_world in two groups
group_rank = rank // 2
color = rank % 2
new_comm = comm.Split(color, group_rank)
# new_comm = comm.Split(color=color)
new_rank = new_comm.Get_rank()
new_size = new_comm.Get_size()

def poisson(alpha):
    mesh = UnitSquareMesh(new_comm, 256, 256)
    V = FunctionSpace(mesh, 'CG', 1)
    u, v = TrialFunction(V), TestFunction(V)
    bc = DirichletBC(V, Constant(0), 'on_boundary')
    a = inner(Constant(alpha)*grad(u), grad(v))*dx
    L = inner(Constant(1), v)*dx
    uh = Function(V)
    solve(a == L, uh, bc)
    info('alpha %g -> |uh|=%g' % (alpha, uh.vector().norm('l2')))
    
alphas = [3, 8]
# Get alpha based on the rank
my_alpha = alphas[new_rank]
poisson(my_alpha)
# mpirun -np 4 python3 file.py
2

You want to extract my_alpha based on the color, rather than the rank, otherwise processors in the same group after split end up having different values of alpha.

Modified code

from dolfin import *
# set MPI communicator
comm = MPI.comm_world
size = comm.Get_size()
rank = comm.Get_rank()
# split comm_world in two groups
group_rank = rank // 2
color = rank % 2
new_comm = comm.Split(color, group_rank)
# new_comm = comm.Split(color=color)
new_rank = new_comm.Get_rank()
new_size = new_comm.Get_size()

def poisson(alpha):
    mesh = UnitSquareMesh(new_comm, 256, 256)
    V = FunctionSpace(mesh, 'CG', 1)
    u, v = TrialFunction(V), TestFunction(V)
    bc = DirichletBC(V, Constant(0), 'on_boundary')
    a = inner(Constant(alpha)*grad(u), grad(v))*dx
    L = inner(Constant(1), v)*dx
    uh = Function(V)
    solve(a == L, uh, bc)
    print(f"new_rank = {new_rank}, color={color}, alpha={alpha} -> |uh|=%g {uh.vector().norm('l2')}")

alphas = [3, 8]
# Get alpha based on the color
my_alpha = alphas[color]
poisson(my_alpha)
# mpirun -np 4 python3 file.py

results in

new_rank = 0, color=1, alpha=8 -> |uh|=%g 1.3203509408791925
new_rank = 1, color=1, alpha=8 -> |uh|=%g 1.3203509408791925
new_rank = 0, color=0, alpha=3 -> |uh|=%g 3.5209358423440125
new_rank = 1, color=0, alpha=3 -> |uh|=%g 3.5209358423440125

For future posts, make sure to include the imports in your snippet.

2 Likes
3

It worked as I wished.
Thank you very much.