Internal:CDT3D SU2: Difference between revisions
| Line 27: | Line 27: | ||
Time required for the first mesh (from Becket) : | Time required for the first mesh (from Becket) : | ||
<blockquote> | <blockquote> | ||
''I just checked the record: for 15 CTU in the periodic steady state region (where we compute mean flow and turbulent stats), it took 3 days and 7 hours on 480 processors. The transient is about another 10 CTUs and that took roughly another 2 days. | ''I just checked the record: for 15 CTU in the periodic steady state region (where we compute mean flow and turbulent stats), it took 3 days and 7 hours on 480 processors. The transient is about another 10 CTUs and that took roughly another 2 days. '' | ||
So in total: 25 CTUs for about 5 days. We are computing longer than the Go4Hybrid guys tho: they use 7+13=20CTU.'' | ''So in total: 25 CTUs for about 5 days. We are computing longer than the Go4Hybrid guys tho: they use 7+13=20CTU.'' | ||
</blockquote> | </blockquote> | ||
Revision as of 10:36, 11 February 2019
General
- Oct 30 2018 presentation [1]
- Pointwise and SU2 [2]
- Paper : Construction of Prism and Hex Layers from Anisotropic Tetrahedra [3] see also [4]
Interesting points from SU2 developers
- linear scalability is expected if there are at least 5000 grid points per core. [5]
BL discussion with Beckett
For the target method RANS+LES boundary layer is needed to resolve the turbulance. Specifically for the Delta wing see thje link bellow for more details of the simulation. During the fine-tunning of the mesh it it expected to require changes mostly in the boundary layer, since, this is where the turbulance near the boundary ( which we are trying to resolve) lives.
Boundary layer is usually given the following restrictions:
- Δs: Distance of first cell from the solid surface
- Thickness of the boundary layer
- Stretching factor / number of layers
More Links
- Open source tool for creating meshes https://www.salome-platform.org/
- European Knowledge base for CFD simulations [6]
- Project page in the site above for the Delta wing case [7]
Delta Wing Project AIAA Aviation 2019
Time required for the first mesh (from Becket) :
I just checked the record: for 15 CTU in the periodic steady state region (where we compute mean flow and turbulent stats), it took 3 days and 7 hours on 480 processors. The transient is about another 10 CTUs and that took roughly another 2 days.
So in total: 25 CTUs for about 5 days. We are computing longer than the Go4Hybrid guys tho: they use 7+13=20CTU.