Hi Etienne, as Christoph points out, the effect you see is that a problem occurs on one process and other processes are waiting. I don't expect the linear solver to cause the deadlock, since quite some care is taken there to avoid exactly that. I recommend that you recompile with debug options. This might give you a more meaningful error message. If not, you have to enter the joy of parallel debugging.
It is very unusual that the AMG is so much slower. As Christoph says, solving directly on the coarsest level can be important. You can try to make a minimal example on top of a standard Dumux model so that we can take a look. Kind regards Bernd Tue, 13 Dec 2016 16:09:25 +0100 Christoph Grüninger <[email protected]> wrote: >Hi Etienne, >as I did never parallel runs with DuMuX myself, I can only guess. > >My guess is, that your linear solver does not converge and you don't see >the error because some process is waiting for the result from a process >that terminated due to the failed convergence. So it will never >show the error or terminate. > >> For the AMG solver, are there some additional parameters to give? > >Do you use SuperLU or UMFPack or an iterative method after you >preconditioned with AMG? Markus Blatt often advises to use the direct >methods as the iterative ones are prone to breakdown. Might be worth a try… > >Bye >Christoph > >-- >AddressSanitizer is not expected to produce false positives. >If you see one, look again; most likely it is a true positive! >[http://clang.llvm.org/docs/AddressSanitizer.html] > _______________________________________________________________ Bernd Flemisch phone: +49 711 685 69162 IWS, Universitaet Stuttgart fax: +49 711 685 67020 Pfaffenwaldring 61 email: [email protected] D-70569 Stuttgart url: www.hydrosys.uni-stuttgart.de _______________________________________________________________ _______________________________________________ Dumux mailing list [email protected] https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
