Hello Wolfgang I made example code for both distributed and fullydistributed triangulations.
https://codeberg.org/cpraveen/deal_ii/src/branch/master/add_periodicity On a 128^3, I get this timing mpirun -np 2 ./main *distributed* +---------------------------------------------+------------+------------+ | Total wallclock time elapsed since start | 463s | | | | | | | Section | no. calls | wall time | % of total | +---------------------------------+-----------+------------+------------+ | Add periodicity | 1 | 445s | 96% | | Collect faces x | 1 | 0.0425s | 0% | | Collect faces y | 1 | 0.0313s | 0% | | Collect faces z | 1 | 0.0335s | 0% | +---------------------------------+-----------+------------+------------+ *fullydistributed* +---------------------------------------------+------------+------------+ | Total wallclock time elapsed since start | 12.2s | | | | | | | Section | no. calls | wall time | % of total | +---------------------------------+-----------+------------+------------+ | Add periodicity | 1 | 0.0038s | 0% | | Collect faces x | 1 | 0.368s | 3% | | Collect faces y | 1 | 0.0153s | 0.13% | | Collect faces z | 1 | 0.0133s | 0.11% | +---------------------------------+-----------+------------+------------+ The major time is taken in add_periodicity function. fullydistributed does this faster, I have to see if other parts of my code work with this triangulation. best praveen On Thu, Dec 18, 2025 at 10:04 AM 'Wolfgang Bangerth' via deal.II User Group <[email protected]> wrote: > On 12/16/25 03:10, Praveen C wrote: > > > > *Case 1.* Suppose I create a 256^3 mesh like this > > > > GridGenerator::subdivided_hyper_rectangle > > Call for x,y,z > > GridTools::collect_periodic_faces > > triangulation.add_periodicity > > > > then code runs for a very long time to do this, and on some machines > crashes > > with oom_kill error, which is related to out of memory error. > > Praveen: > Can you be more specific? Show us a minimal code that demonstrates the > problem > and that we can run. It should be possible to do this with not much more > than > 20 lines of code. > > Specifically, in which operation is the CPU time used, and in which do you > run > out of memory? > > (My best guess is that the algorithms use a double loop over all coarse > mesh > cells. This works well if you have a few dozen or a few hundred coarse > mesh > cells, as we often do. But your 195x72x106 mesh has 1.5M coarse mesh > cells, > and I'm not surprised that that doesn't work. You won't fix that issue by > going to a different triangulation class. I think it's inherent in the > current > design of the periodic face finding algorithm, but we can't know for sure > unless we know where exactly the problem appears.) > > Best > W. > > -- > The deal.II project is located at http://www.dealii.org/ > For mailing list/forum options, see > https://groups.google.com/d/forum/dealii?hl=en > --- > You received this message because you are subscribed to the Google Groups > "deal.II User Group" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To view this discussion visit > https://groups.google.com/d/msgid/dealii/547e683a-8075-49cc-8f09-6f59a0ae04bd%40colostate.edu > . > -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/d/msgid/dealii/CAEvUdMJDy2ZSd4Mj9vS%2BLtdwHk8LyiaV_kLtzRjjx7SJ3JSECw%40mail.gmail.com.
