next up previous contents
Next: 3D Spray Up: Benchmarking Previous: Dam Break Tutorial   Contents

3D Dam Break

With the assistance of OpenCFD [4], we then extended the 2D Dam Break tutorial case to run in full 3D, with such a fine mesh that water droplets can be seen.

The value of hex in blockMeshDict is set to

hex (0 1 2 3 4 5 6 7) (184 184 184) simpleGrading (1 1 1)
which gives 6.23 million cells.

Interestingly, as part of the pre-processing routines, the routine blockMesh failed due to lack of memory when running in HECToR's serial queue. This was circumvented by running blockMesh within a small parallel queue. (See Section 4.3).

Figure: Time (secs) for the 3D Dam Break case, where `damBreak' is the execution times and `damBreak.lin' is the line of perfect scaling
\includegraphics[width=10cm]{dambreak.png}


Table 3: Timing,and performance results and processor topology employed or 3D Dam Break case with 8.23 million cells
Number 3D Dam Break processor
of cores Time (Perf) topology
8 688.3 (-) 4 x 2 x 1
16 362.4 (1.90) 4 x 4 x 1
32 183.8 (1.97) 4 x 4 x 2
64 94.0 (1.96) 4 x 4 x 4
128 53.6 (1.75) 8 x 4 x 4
256 42.3 (1.27) 8 x 8 x 4
512 38.1 (1.11) 8 x 8 x 8

It can be seen from table 3 that the optimum number of processors is 128, when running interFoam using 6.23 million cells.

The case was configured so that increasing the resolution was simply updating the value of hex in blockMeshDict to read

hex (0 1 2 3 4 5 6 7) (230 230 230) simpleGrading (1 1 1)
which gives 12.17 million cells. However, it was found that this case was simply too large to pre-process on either HECToR or Ness.

The 3D Dam Break benchmark case is available on the HECToR's OpenFOAM web page [12].


next up previous contents
Next: 3D Spray Up: Benchmarking Previous: Dam Break Tutorial   Contents
Gavin J Pringle
2010-04-16