Well if some nodes are at OpenMPI 1.4 - and others are at OpenMPI 1.3.4 - that might explain the problem :) Satish On Thu, 17 Dec 2009, Kevin.Buckley at ecs.vuw.ac.nz wrote: > >> =========== > >> > >> My PETSc was built against OpenMPI 1.4 > >> > > > > Apologies, > > > > That should have read: OpenMPI 1.3.4 > > Actually, I was right first time. > > OpenMPI 1.4 it is ! > >