I see, What version of MPI are you using? Some versions of OpenMPI have serious memory leaks. Maybe try the exact same version of OpenMPI you used previously with success?
You can also use valgrind with some options to track the memory usage. Barry > On Dec 24, 2019, at 10:03 AM, Marius Buerkle <[email protected]> wrote: > > I see. For actual production runs on 512 processes with larger matrices this > causes eventually a seg fault due to running out of memory. I am not sure but > think this did not happen with some "older" version of PETSC. I am not sure > which one (or which revision), I wil check it. > > > Gesendet: Mittwoch, 25. Dezember 2019 um 00:53 Uhr > Von: "Smith, Barry F." <[email protected]> > An: "Marius Buerkle" <[email protected]> > Cc: "Mark Adams" <[email protected]>, "petsc-usersmcs.anl.gov" > <[email protected]> > Betreff: Re: [petsc-users] possible memory leak > > There are no leaks but it appears what is happening is that rather than > recycle the memory PETSc is returning to the system the system is generating > new space as needed. Since the old PETSc pages are never used again this > should be harmless. > > Barry > > > > > On Dec 24, 2019, at 9:47 AM, Marius Buerkle <[email protected]> wrote: > > > > thanks for the swift reply. This didn't give any output. > > > > > > Gesendet: Dienstag, 24. Dezember 2019 um 23:56 Uhr > > Von: "Mark Adams" <[email protected]> > > An: "Marius Buerkle" <[email protected]> > > Cc: "petsc-usersmcs.anl.gov" <[email protected]> > > Betreff: Re: [petsc-users] possible memory leak > > Try running with -malloc_debug > > > > This should print out where unfreed memory was allocated. See what pops up. > > > > Mark > > > > On Tue, Dec 24, 2019 at 2:27 AM Marius Buerkle <[email protected]> wrote: > > Hi, > > > > > > In my code I create and destroy during one run a lot of MPIAIJ matrices and > > it seems that not all memory is freed when I destroy the matrices. > > Moreover, it looks like the amount of “disappearing” memory increases with > > the number of processes. I am not so sure if I am doing something wrong or > > if it is really a memory leak. I have attached a simple reproducer. > > > > > > Best, > > > > Marius > > >
