Poisson testcase in parallel on cluster
Dear All,
I'm running the Dolfin script solving the poisson equation on a cluster to test FEniCS parallel performance (the only change I've made with respect to the original script is the number of degrees of freedom and the solver: the mesh is then a box with N=1000 and I use a gmres solver for memory reasons). I encounter a strange behaviour when I run the test on more than 2 nodes. In fact, when I run the case on 4 nodes and 32 cores, the simulation executes all the commands, without exiting though. Some process continues to run and I need to cancel the job manually. If I request 8 nodes instead, the simulation hangs forever, without executing all the commands, and again it needs to be cancelled manually.
I attach the poisson_demo file, together with the launching script.
