Nice catch, thanks, no, I overlooked the extra options when reverse engineering the test from the makefiles. Now I get identical output.
Thanks, Dominik On Mon, Dec 5, 2011 at 3:24 PM, Matthew Knepley <knepley at gmail.com> wrote: > On Mon, Dec 5, 2011 at 8:17 AM, Dominik Szczerba <dominik at itis.ethz.ch> > wrote: >> >> I am on a Cray system, where I have configured and built petsc 3.2 p5 >> myself, only using the vendor provided mpi, all the other packages >> (parmetis, hypre, mumps) are built with --download-package. "Make ... >> test" does not work here to verify the correct build because users do >> not have access to mpiexec. So I do the test by hand, and I am unsure >> about what I am getting. When executing the test ex19 with the >> system's scheduler I get this output: >> >> lid velocity = 0.0204082, prandtl # = 1, grashof # = 1 >> Number of Newton iterations = 2 >> lid velocity = 0.0204082, prandtl # = 1, grashof # = 1 >> Number of Newton iterations = 2 >> >> It's quite different from the expected output in ex19_1.testout: >> >> lid velocity = 0.0016, prandtl # = 1, grashof # = 1 >> Number of Newton iterations = 2 >> lid velocity = 0.0016, prandtl # = 1, grashof # = 1 >> Number of Newton iterations = 2 > > > Are you running with the right options???-dmmg_nlevels 4 -snes_monitor_short > > ? Matt > >> >> Does it mean that I have screwed up the build somehow, or is this >> somehow the effect of the batching system, or just numerical >> uncertainty due to the parallel computation? Note: I get the same >> result with 1 and a few MPI processes. >> >> Many thanks for any ideas. >> >> Dominik > > > > > -- > What most experimenters take for granted before they begin their experiments > is infinitely more interesting than any results to which their experiments > lead. > -- Norbert Wiener
