[MOAB-dev] problems reading meshes in MOAB in the 'read_part' mode

Jason Kraftcheck kraftche at cae.wisc.edu
Thu Apr 1 16:00:25 CDT 2010


Dmitry Karpeev wrote:
> I guess it was somewhat strong, but what I really meant is this:
> "For whatever reason I can't seem to make 'read_part' work with any mesh
> or any machine with any number of procs I have tried recently, so
> for the purpose of my task at hand -- benchmarking of parallel reads --
> I will avoid using 'read_part', at least for now."
> 
> The reason I brought up the various reported errors was to marshal evidence
> that the problem is probably with 'read_part' itself -- not the meshes
> I feed it.
> 

Well, this particular problem was definitely with the mesh that fed it.

As for read_part in general, I it appears to work for me on both my local
machine and cosmea.  Reading a file containing 64 blocks of 1000 hexes each
worked fine for 4 to 32 processes (64-proc is still waiting in queue on
Cosmea.)  My input was generated in cubit and converted to h5m using
mbconvert before running the parallel test.  The cubit model was a 4x4x4
grid of 10x10x10 bricks, merged, meshed with size 1, with one exodus block
per brick.  I then tested the parallel read with:
  ~/submit -n 32 ./mbconvert -O PARALLEL=READ_PART \
      -O PARTITION=MATERIAL_SET -f VTK blocks-64.h5m /dev/null

- jason


More information about the moab-dev mailing list