GEOM vs OBST case

[fidran]

FDS-6.9.1-0-g889da6a-release
Attached is a case where the building has a cylindrical shape. I made a simplified model of the atrium inside the building with Blender and exported it as an OBST case and a GEOM case. There is a big different in simulation time between the OBST case and the Geom case (1:5). Also, in memory consumption there is a significant difference 1:4. From the OUT fil I can see the FDS uses different pressure solvers for the OBST case and the GEOM case. FFT vs ULMAT. I tried to change from ULMAT to FFT (&PRES SOLVER='FFT' /), but this line seems to be ignored. Is there a way I can speed up the GEOM case ?

Atrium.zip

[mcgratta]

How many meshes are you using?

[fidran]

One mesh i both cases

&MESH ID='New MESH' IJK=136,136,44 XB=-17.000,17.000,-17.000,17.000,0.000,11.000, MPI_PROCESS=0 / Cell Qty: 813824

[mcgratta]

That's the problem. GEOM requires a different pressure solver that takes a large amount of memory on large meshes. Try breaking this case up into 9 or 16 meshes.

[fidran]

Managed to reduce the difference in computation time (OBST/GEOM) from 1:5,5 to 1:2,5 by splitting the cases into more meshes. It seems that the number of meshes affects the GEOM case more than the OBS case. Thanks for helping.

[marcosvanella]

Hi @fidran, correct, the default Pressure solver when geoms are present is ULMAT, which uses Pardiso currently. This is a linear solver that does an LU decomposition of the discretization matrix. Its cost and memory use increases rapidly with the mesh size. Up to ~50^3 cells per mesh is recommended.We cannot use the default FFT solver for pressure because we don't have a structured discretization around geoms.
Can you send me your test case? I'd like to do some tests with it.
Thanks for reporting.

[fidran]

The result 1:5,5 is not correct. I discovered that the case has been running on a USB disk.
Some new test results are written at the end of the MESH lines. Cases uploaded.
NiST(2).zip

[marcosvanella]

Thank you @fidran!