Re: [DuMux] Switching linear solver causes simulation running on just a part of a 2D grid
> On 2. Sep 2020, at 18:39, Dmitry Pavlov wrote: > > Timo, > > Thank you. I am exploring my options. Are all three steps (Assemble, Solve, > Update) normally parallelized? yes > > I also noticed that AMG does itself use UMFPack as a coarse solver, it it > (well, SuiteSparse) is installed in the system. It does so even in parallel > mode, so maybe UMFPack can be useful in parallel mode after all. > > It is interesting that UMFPack is probably considered as an unconditionally > good thing in AMG. Whenever it is installed, it is used, with the only way to > override being #define DISABLE_AMG_DIRECTSOLVER 1. > UMFPack is used as the coarse grid solver in the AMG preconditioner. As the "coarse grid" is usually quite small a direct solver can give you better performance there. And depending on the accumulation the coarse grid solve will be executed on a single process and the result will be broadcasted to the rest. So it’s used in the parallel solver but runs only on a single process. Timo > Best regards, > > Dmitry > > > >> >> Hi Dmitry, >> >> you can run ILU0BiCGSTAB in parallel yes. UMFPack doesn’t have a parallel >> interface in Dune/Dumux. >> You can choose from the different iterative preconditions and solvers from >> dune-istl. >> >> Timo >> >> >> > ___ > DuMux mailing list > DuMux@listserv.uni-stuttgart.de > https://listserv.uni-stuttgart.de/mailman/listinfo/dumux ___ DuMux mailing list DuMux@listserv.uni-stuttgart.de https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
Re: [DuMux] Switching linear solver causes simulation running on just a part of a 2D grid
Timo, Thank you. I am exploring my options. Are all three steps (Assemble, Solve, Update) normally parallelized? I also noticed that AMG does itself use UMFPack as a coarse solver, it it (well, SuiteSparse) is installed in the system. It does so even in parallel mode, so maybe UMFPack can be useful in parallel mode after all. It is interesting that UMFPack is probably considered as an unconditionally good thing in AMG. Whenever it is installed, it is used, with the only way to override being #define DISABLE_AMG_DIRECTSOLVER 1. Best regards, Dmitry Hi Dmitry, you can run ILU0BiCGSTAB in parallel yes. UMFPack doesn’t have a parallel interface in Dune/Dumux. You can choose from the different iterative preconditions and solvers from dune-istl. Timo ___ DuMux mailing list DuMux@listserv.uni-stuttgart.de https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
Re: [DuMux] Switching linear solver causes simulation running on just a part of a 2D grid
-- _ Timo Koch phone: +49 711 685 64676 IWS, Universität Stuttgart fax: +49 711 685 60430 Pfaffenwaldring 61 email: timo.k...@iws.uni-stuttgart.de D-70569 Stuttgart url: www.iws.uni-stuttgart.de/en/lh2/ _ > On 2. Sep 2020, at 13:59, Dmitry Pavlov wrote: > > Hi Timo, > > After some time on DuMux 3.1 and AMGBackend, I have moved to DuMux 3.2 (and > thus AMGBiCGSTABBackend, since the AMGBackend is deprecated). > > I would like to give a try now to ILU0BiCGSTAB and/or UMFPack solvers, if it > is possible to run them in parallel. Is the information that you provided in > July (see below) relevant now? Are those solvers available in parallel mode > in this development branch? Hi Dmitry, you can run ILU0BiCGSTAB in parallel yes. UMFPack doesn’t have a parallel interface in Dune/Dumux. You can choose from the different iterative preconditions and solvers from dune-istl. Timo > > Best regards, > > Dmitry > > > > On 14.07.2020 13:45, Timo Koch wrote: >> Hi Dmitry, >> >> assuming, as Bernd suggested, that you run in parallel the solvers you tried >> won’t work since they are sequential solvers. >> You can currently use the ISTL solver factory with the dune modules on the >> master branch to try other solvers in parallel. >> There is also a development branch "feature/new-istl-linear-solvers” (see >> merge request !2113) that will make it easier to create parallel solvers >> other than the AMGBackend. >> >> For the ISTL solver factory you have a look at the shallow water dambreak >> test which is tested in parallel if a recent dune-istl version is available >> (https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/tree/master/test/freeflow/shallowwater/dambreak). >> The solvers are configured in the input file. (It takes longer to compile >> since all solvers are choosable at runtime.) >> >> Timo >> >> ___ DuMux mailing list DuMux@listserv.uni-stuttgart.de https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
Re: [DuMux] Switching linear solver causes simulation running on just a part of a 2D grid
Hi Timo, After some time on DuMux 3.1 and AMGBackend, I have moved to DuMux 3.2 (and thus AMGBiCGSTABBackend, since the AMGBackend is deprecated). I would like to give a try now to ILU0BiCGSTAB and/or UMFPack solvers, if it is possible to run them in parallel. Is the information that you provided in July (see below) relevant now? Are those solvers available in parallel mode in this development branch? Best regards, Dmitry On 14.07.2020 13:45, Timo Koch wrote: Hi Dmitry, assuming, as Bernd suggested, that you run in parallel the solvers you tried won’t work since they are sequential solvers. You can currently use the ISTL solver factory with the dune modules on the master branch to try other solvers in parallel. There is also a development branch "feature/new-istl-linear-solvers” (see merge request !2113) that will make it easier to create parallel solvers other than the AMGBackend. For the ISTL solver factory you have a look at the shallow water dambreak test which is tested in parallel if a recent dune-istl version is available (https://git.iws.uni-stuttgart.de/dumux-repositories/dumux/-/tree/master/test/freeflow/shallowwater/dambreak). The solvers are configured in the input file. (It takes longer to compile since all solvers are choosable at runtime.) Timo ___ DuMux mailing list DuMux@listserv.uni-stuttgart.de https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
