Is there a way to determine the amount of unknowns, i.e. for example the amount of grid points?
Concerning your second point: Yes, only when doing a mesh refinement the memory consumption increases, else not. Am Dienstag, 3. Oktober 2017 15:08:19 UTC+2 schrieb Wolfgang Bangerth: > > On 10/03/2017 01:04 AM, 'Maxi Miller' via deal.II User Group wrote: > > I have a system of coupled heat equations, written for MPI using > > > > | LinearAlgebraTrilinos::MPI::SparseMatrixsystem_matrix; | > > > > as the sparse system matrix. Now for certain values the program runs > fine, > > but for other values (which I am interested in) the RAM consumption goes > up > > rapidly after several steps in time and refinement. > > > > Valgrind tells me that the biggest reason for that is the system matrix. > > This makes the program unusable workstations with 32 GByte RAM. Is that > > something I could expect, or is that a bug in my program? > > It's likely a bug of some sort, unless you are solving problems with LOTS > of > unknowns. A general rule is that you need maybe 1500 or 2000 bytes per > unknown, so you should be able to solve problems with several million > unknowns > with 32 GB. How large are the problems you have? > > > > I am solving the equations with the method shown in step-15 and step-33, > > i.e. I iterate over an inner loop for every time step until I am below a > > certain l2-value, and then go to the next time point. > > > > One possible approach I could think of is to set the threshold higher, > but > > that would only delay the problem. Another approach would be to check if > > the l2-value did not change over the last five internal loops, > > > > and then continue anyway, but that would again increase the error. Is > there > > a better way? > > That doesn't make any sense to me. The only thing that should increase the > memory consumption should be mesh refinement. The number of iterations > should > have no effect. If it does, then you have a memory leak somewhere. > > Best > W. > > -- > ------------------------------------------------------------------------ > Wolfgang Bangerth email: bang...@colostate.edu > <javascript:> > www: http://www.math.colostate.edu/~bangerth/ > > -- 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 dealii+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
