On Sat, 12 Dec 2009, Przemyslaw Kaminski wrote: > Thank you for your answer. The problem is that when I run ex9, say, then > it actually takes twice as long for two processors. I do: > mpiexec -np 2 ./ex9-opt > It takes 7.2 sec, just running ./ex9-opt takes 3.4 sec.
On my machine with my default solver options, "mpirun -np 4 ./ex9-opt -other_options" takes 5.2 seconds; "-np 2" takes 6.3 seconds; "-np 1" takes 9.9 seconds. Not exactly linear speedup, but you wouldn't expect linear speedup on such a small problem. Just setting up and tearing down the problem takes longer on 2 processors than on one. > I think the program is just run twice, not with two threads which > should communicate with each other... I have seen that happen before, when using one MPI implementation to run a libMesh binary that was compiled with a different MPI (or with no MPI at all). ex9 should print out a line like "n_processors() = 2" right when it starts up; if you're seeing "n_processors() = 1" instead then you need to make sure that your libMesh install configured correctly to use your mpiexec. --- Roy ------------------------------------------------------------------------------ Return on Information: Google Enterprise Search pays you back Get the facts. http://p.sf.net/sfu/google-dev2dev _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
