Hi Shawn, Nice to hear that compilation was a piece of cake. (Historically it was a major pain in the ass.)
Partitioning the grid is pretty easy in DUNE: Just call grid.loadBalance() and the rest is done automatically, provided that you your MPI impelmentation is compatible with DUNE and you correctly set up the MPI at your main function (using Dune::MpiHelper). Note that some grids (e.g. Dune::YaspGrid) require the MPI communicator passed to their constructor. If you use the latest Dumux (release 2.1 or the trunk versions), all of this is taken care of by the Dumux::start() function, though. Please also note that last week we've discovered out that our parallel linear solver code has a bug which causes it to not work with overlaps other than 1. As a stop-gap measure, the overlap was thus reduced to 1 just before the release. If you use an older version, you should upgrade and try again. have fun Andreas On Montag, 5. März 2012 you wrote: > Hi Andreas and All, > > I finally get around and try the parallel impelmentation. The compiling > part is very straightforward. However, when I run it (e.g., > boxmodels/test_1p), it appears that the grid need to be somehow > partitioned in order for the code to work in parallel. Do we have any > documentation on how to achieve this? > One test, boxmodels/test_2p2cni, appears to be running in parallel (two > processors I specified are both 100% busy), however, it appears to be in > an infinite loop, dumping a lot of data on the disk but complaining > iterative solver not converging. > Do we have an example that will run correctly in parallel? > > Also, are there any improvement on the latest release 2.1? > > Sorry about all the questions but I am a newbie to this fairly complex > system. > Thanks for your attention and help. > > -Shawn > > From: Shawn Zhang <[email protected]> > To: Andreas Lauser <[email protected]>; DuMuX User Forum > <[email protected]> Sent: Thursday, February 16, 2012 11:07 > AM > Subject: Re: [DuMuX] Parallel execution of Dumux examples > > > Hi Andreas, > > Thanks a lot for your attention and valuble instruction. > > To make sure I understand the process, I would like to have one quick > clarification. The parallel capability does not require any modification > of code at the Dumux application level, is this correct? In another world, > domain decomposition and other parallel specific tasks are handled in a > way that is transparently to application code, right? > Best regards, > -Shanw > > From: Andreas Lauser <[email protected]> > To: Shawn Zhang <[email protected]>; DuMuX User Forum > <[email protected]> Sent: Thursday, February 16, 2012 9:24 > AM > Subject: Re: [DuMuX] Parallel execution of Dumux examples > > Hi Shawn, > > First, some caveats: the parallel solvers in Dumux are currently still > experimental, and may eat kids. (i.e. they may lead to invalid results, > simulation aborts or they may be slower than the sequential solvers.) The > problem with them is that the linear solvers provided by DUNE-ISTL don't > work to well in parallel yet. If you want to give it a try anyway, read on > and send us your patches! > > To use Dumux with MPI, you first have to install an MPI library with the > header files on your machine. (personally, I use openmpi 1.4.5, but it > also there are rumors that it also works with other MPI implementations > like MPIch2.) > > Then, you need to compile DUNE and Dumux with the configure option > --enable- parallel (usually you specify this by means of an .opts file for > dunecontrol): > > ./dune-common/bin/dunecontrol --opts=$YOUR_OPTS_FILE --module=dumux all > > then, you need to compile the Dumux application, you want to run in > parallel. For the two-phase example this is achieved by > > cd dumux/test/boxmodels/2p > make check > > and finally you have to execute the resulting binary using mpirun: > > mpirun -np $NUM_PROCESSORS ./test_2p $OPTIONS_LIKE_FOR_SEQUENTIAL_RUN > > Note, that if you need non-cartesian grids for you simulation, you should > also install ALUgrid, see > > http://aam.mathematik.uni-freiburg.de/IAM/Research/alugrid/ > > for details. > > Since the Dumux SVN repository currently is in hard feature freeze to > prepare for the next release, we would appreciate if you report your > experience if you intend to use the latest SVN version. > > good luck and have fun > Andreas > > On Dienstag, 14. Februar 2012 you wrote: > > Dear Dumux Developers, > > > > > > > > I want to run my Dumux programs in parallel on a multi-core workstation. > > As an example, I want to test run the tutorial_coupled example in > > parallel mode. > > Can you kindly provide me some pointers to documents or codes in order to > > implement this? Any suggestion will be truly appreciated, as I am fairly > > new to the Dumux environment. > > Thank you very much for your attention. > > > > > > > > -Shawn > > +1-978-494-3016 -- Andreas Lauser Department of Hydromechanics and Modelling of Hydrosystems University of Stuttgart Pfaffenwaldring 61 D-70569 Stuttgart Phone: (+49) 711 685-64719 Fax: (+49) 711 685-60430 www.hydrosys.uni-stuttgart.de _______________________________________________ Dumux mailing list [email protected] https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
