There's an assumption that all the processors are working on the same problem so each processor gets a part of the mesh. Are each of the meshes truly independent so they don't share any DOFs? If they really are independent have you considered just running thousands of jobs instead?
Cody On Mon Feb 09 2015 at 10:59:49 AM Peter Wijeratne <[email protected]> wrote: > Hello libMesh community, > > Bit of a special (as opposed to general) problem here. I'm trying to > create a serial mesh on each of N processors, each to be kept on its > respective processor and solved separately. This causes problems when > initialising a given mesh's EquationSystems; I've narrowed the bug down > to this line of code in dof_map.C: > > Parallel::send_receive(procup, requested_ids[procup], > procdown, request_to_fill); > > For the case where N = 2, when run on the first processor the above > request_to_fill vector has a size equal to the number of nodes (or > elements) on the second processor. I think this is where the problem > occurs --- despite localising all the nodes and elements of the first > processor's mesh, because I've already got a completely different mesh > on the second processor it returns the number of nodes for that mesh. > This either causes a crash if n_nodes_mesh_1 > n_nodes_mesh_0, or funky > behaviour otherwise. > > Is there a workaround for this? I understand that this is an unusual > problem, but ultimately I aim to run several thousand serial meshes so > parallelisation is pretty important. > > Thanks in advance, > Peter > > ------------------------------------------------------------ > ------------------ > Dive into the World of Parallel Programming. The Go Parallel Website, > sponsored by Intel and developed in partnership with Slashdot Media, is > your > hub for all things parallel software development, from weekly thought > leadership blogs to news, videos, case studies, tutorials and more. Take a > look and join the conversation now. http://goparallel.sourceforge.net/ > _______________________________________________ > Libmesh-users mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/libmesh-users > ------------------------------------------------------------------------------ Dive into the World of Parallel Programming. The Go Parallel Website, sponsored by Intel and developed in partnership with Slashdot Media, is your hub for all things parallel software development, from weekly thought leadership blogs to news, videos, case studies, tutorials and more. Take a look and join the conversation now. http://goparallel.sourceforge.net/ _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
