[petsc-users] memory increase in a transient calculation but no valgrind errors

[Dominik Szczerba]

On Sat, Dec 24, 2011 at 9:13 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> On Sat, Dec 24, 2011 at 14:10, Dominik Szczerba <dominik at itis.ethz.ch>
> wrote:
>>
>> I do not understand: is there something in Petsc of this sort that I
>> could use for my unstructured meshes or you meant I have to implement
>> it myself?
>
>
> If you implement DM to the extent of providing DMCreateGlobalVector(), then
> DMGetGlobalVector()/DMRestoreGlobalVector() will work for you. Otherwise you
> implement it yourself. This kind of caching is a very common pattern for
> data structure reuse.

I see. I guess it's out of reach for the moment. Latest news: at some
point the linearly growing memory usage line starts to saturate. Would
it not confirm your guess about fragmentation.

Thanks, Dominik