[petsc-users] GAMG speed

[Barry Smith]

On Aug 13, 2013, at 9:57 PM, Michele Rosso <mrosso at uci.edu> wrote:

> Hi Jed,
> 
> I attached the output for both the runs you suggested. At the beginning of each file I included the options I used.
> 
> On a side note, I tried to run with a grid of 256^3 (exactly as before) but with less levels, i.e. 3 instead of 4 or 5.
> My system stops the run because of an Out Of Memory condition. It is really odd since I have not changed anything except
> - pc_mg_levels.  I cannot send you any output since there is none. Do you have any guess where the problem comes from? 

   By default it uses a direct solver (maybe even sequential) for the coarsest level; since the coarse level is big the direct solver requires too much memory.  You could install PETSc with the ./configure option --download-superlu_dist and run with -mg_coarse_pc_type lu -mg_coarse_pc_factor_mat_solver_package superlu_dist

  Barry


> Thanks,
> 
> Michele
> 
> On 08/13/2013 07:23 PM, Jed Brown wrote:
>> Michele Rosso <mrosso at uci.edu>
>>  writes:
>> 
>>> The matrix arises from discretization of the Poisson equation in 
>>> incompressible flow calculations.
>>> 
>> Can you try the two runs below and send -log_summary?
>> 
>>   -log_summary -ksp_monitor -ksp_view -ksp_converged_reason -pc_type mg -pc_mg_galerkin -pc_mg_levels 5 -mg_levels_ksp_type richardson -mg_levels_ksp_max_it 1
>> 
>> 
>>   -log_summary -ksp_monitor -ksp_view -ksp_converged_reason -pc_type mg -pc_mg_galerkin -pc_mg_levels 5 -mg_levels_ksp_type richardson -mg_levels_ksp_max_it 1 -pc_mg_type full
>> 
> 
> <output_mg_1.txt><output_mg_2.txt>