[petsc-users] dmmg_grid_sequence

[Matthew Knepley]

On Mon, Apr 2, 2012 at 11:33 AM, Xuefeng Li <li at loyno.edu> wrote:

> On Mon, 2 Apr 2012, Matthew Knepley wrote:
>
>  On Mon, Apr 2, 2012 at 10:11 AM, Xuefeng Li <li at loyno.edu> wrote:
>>
>>  Hi, all.
>>>
>>> I am using multigrid with DMMG (5 levels) in my petsc program.
>>> Turning on the dmmg_grid_sequence option seems
>>> to slow down the program dramatically. For instance,
>>> the program converges in 3 iterations in 6.270e+01 sec
>>> when dmmg_grid_sequence is OFF; it also converges
>>> in 3 iterations but in 1.699e+04 sec when dmmg_grid_sequence
>>> is ON.
>>>
>>> Is this typical of petsc (I'm using petsc-3.1-p8)?
>>> Are there ways to improve the performance when
>>> dmmg_grid_sequence is ON?
>>>
>>>
>> Its impossible to know what is happening from this description. There is
>> no
>> "generic" performance of grid sequencing. The best thing I can say to this
>> is,
>> it looks like the interpolated solution to your coarse problem is not a
>> great
>> solution to your fine problem.
>>
>>  That is possible.
>
>
>
>  However, you could be oversolving the coarse
>> problems, blowing up the work, or interpolating badly for your problem.
>>
>>  What do you mean by "oversolving the coarse problems"?
> I am using the same stopping conditions in the two runs.
>

Which may be too tight for the coarse runs.


> The only difference is dmmg_grid_sequence ON and OFF.
>
> I also observed that in other situations, when dmmg_grid_sequence
> is ON, the execution will abort due to "Out of memory" error,
> whereas when OFF, the program converges without error.
> How much more memory does grid-sequence employ? Twice as much?
>

If you use Galerkin MG, then t would depend on how full your Jacobian is.

   Matt


> Regards,
>
> --Xuefeng Li, (504)865-3340(phone)
>  Like floating clouds, the heart rests easy
>  Like flowing water, the spirit stays free
>       http://www.loyno.edu/~li/home
>  New Orleans, Louisiana (504)865-2051(fax)
>
>


-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120402/b6ef910c/attachment-0001.htm>