Behaves for me as expected cores | wall-time 1 | 00:11:19 2 | 00:06:19 4 | 00:03:48 8 | 00:02:42
For 8 cores DOF count per process becomes rather small so that communication overhead is large. And sparse LU solvers are not scalable. Jan On Wed, 1 Oct 2014 00:28:51 +0300 Serbulent UNSAL <[email protected]> wrote: > Hi, > > I'm new in fenics and try to figure out parallel capabilities with a > time dependent diffussion equation. I use example at fenics website > with small modifications. > > I run it with "mpirun -np 8 python scriptname.py" for parallel > solution. > > But serial version works faster than parallel version. (00:02:10 for > serial and 00:02:53 for parallel) > > Code could be found below. I'll be thankful for any ideas. > > http://paste.ofcode.org/thrvKAfGymwegMg5TkwpLt > > Thanks, > > Serbulent _______________________________________________ fenics-support mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics-support
