Dear Wolfgang, > I think there's little to gain from trying to tweak the memory allocation part of this. The fact that MUMPS takes a lot of memory is also nothing you can change -- that's just what you get for using a direct solver.
Thanks for your thoughts on this. I have tried to use iterative solvers, but they all seem to fail after at some point, and it seems that the maximum number of iterations in every one of them is capped at 10000. I will give them another try before accepting this RAM devouring fate. Worst case scenario, I hope that when I distribute this to several nodes the "parallel ram" will (even if slightly) compensate this effect. > Nice graph, though. It clearly shows what's going on! Thank you! I hope it makes up for the unclear first post! Bests, L On Thursday, December 7, 2017 at 4:44:46 PM UTC+1, Wolfgang Bangerth wrote: > > > Lucas, > > > I mean that the LU Decomposition solver takes a huge amount of RAM, and > it > > seems to me that allocating that once and reusing the space would be > better. > > Attached you can find a simple graph* showing how the free memory in > time. I > > ran an instance of my program using around 164k cells, running on 7 > threads. > > As you can see, the solving step consumes a lot of RAM, and then > deallocates > > it after the solver finishes. What I wonder is if it is useful and > possible to > > just do this allocation/freeing once, at the start of the program. > > I don't think it matters. First, you can't know how much memory you're > going > to use if you do a sparse LU decomposition. It all depends on the sparsity > pattern of the matrix. Second, MUMPS does this internally -- I don't think > you > have control over it. Third, sparse decompositions are so expensive to > compute > that the cost of memory allocation is likely completely negligible. > > I think there's little to gain from trying to tweak the memory allocation > part > of this. The fact that MUMPS takes a lot of memory is also nothing you can > change -- that's just what you get for using a direct solver. > > Nice graph, though. It clearly shows what's going on! > > Cheers > 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.
