### mpi code for parallel resolution

75

views

0

Dear all,

thank you in advance for the help.

I'm trying to run the following code in parallel on 200 k elements mesh, typing "mpirun -n np python demo.py":

(uu0,pp0) = w0.split(True)

delta1.update(uu0, nu,dt, kappa1, u_in, diameter)

delta2.update(uu0, nu, dt, kappa2, u_in, diameter)

cell1.update(uu0, nu,dt, kappa1, u_in, diameter)

cell2.update(uu0, nu, dt, kappa2, u_in, diameter)

DELTA1= interpolate(delta1, DGeW)

DELTA2=interpolate(delta2, DGeW)

CELL1= interpolate(cell1, DGeW)

CELL2=interpolate(cell2, DGeW)

#Save deltas

Delta1_vtk<< (DELTA1,t)

Delta2_vtk<< (DELTA2,t)

Cell1_vtk<< (CELL1,t)

Cell2_vtk<< (CELL2,t)

#CFL number

h = CellSize(mesh)

CFL = project(sqrt(inner(uu0,uu0))*dt/h, DGeW)

CFL_vtk<< (CFL,t)

J = derivative (F,w,dw)

# Define variational problem

pde = NonlinearVariationalProblem(F,w,bc,J)

solver = NonlinearVariationalSolver(pde)

prm = solver.parameters

prm["newton_solver"]["absolute_tolerance"] = 1E-8

prm["newton_solver"]["relative_tolerance"] = 1E-7

solver.solve()

(u, p) = w.split()

p_vtk<< (p,t)

u_vtk<< (u,t)

Unfortunately we don't see any time advantage, actually we verify an increment for the resolution time when we increase the processor number. Is there something wrong?

Thank you very much again

Best Regards

M Pesarin

thank you in advance for the help.

I'm trying to run the following code in parallel on 200 k elements mesh, typing "mpirun -n np python demo.py":

(uu0,pp0) = w0.split(True)

delta1.update(uu0, nu,dt, kappa1, u_in, diameter)

delta2.update(uu0, nu, dt, kappa2, u_in, diameter)

cell1.update(uu0, nu,dt, kappa1, u_in, diameter)

cell2.update(uu0, nu, dt, kappa2, u_in, diameter)

DELTA1= interpolate(delta1, DGeW)

DELTA2=interpolate(delta2, DGeW)

CELL1= interpolate(cell1, DGeW)

CELL2=interpolate(cell2, DGeW)

#Save deltas

Delta1_vtk<< (DELTA1,t)

Delta2_vtk<< (DELTA2,t)

Cell1_vtk<< (CELL1,t)

Cell2_vtk<< (CELL2,t)

#CFL number

h = CellSize(mesh)

CFL = project(sqrt(inner(uu0,uu0))*dt/h, DGeW)

CFL_vtk<< (CFL,t)

J = derivative (F,w,dw)

# Define variational problem

pde = NonlinearVariationalProblem(F,w,bc,J)

solver = NonlinearVariationalSolver(pde)

prm = solver.parameters

prm["newton_solver"]["absolute_tolerance"] = 1E-8

prm["newton_solver"]["relative_tolerance"] = 1E-7

solver.solve()

(u, p) = w.split()

p_vtk<< (p,t)

u_vtk<< (u,t)

Unfortunately we don't see any time advantage, actually we verify an increment for the resolution time when we increase the processor number. Is there something wrong?

Thank you very much again

Best Regards

M Pesarin

Community: FEniCS Project

Please login to add an answer/comment or follow this question.