[aerosayan]

Hello everyone,

Which grid formats should our solvers support for maximum support, user friendliness, and performance?

Obviously no perfect answer might exist, so I'm just looking for your personal views. Different answer for each criteria is acceptable, as we'll support multiple formats for the need of different users.

Metrics:

• support : almost every grid generator should be able to export to this format. Personally I'm kind of liking the I-deas Universal format (*.unv), as it's old, and both gmsh and salome support it.
• user friendliness : user finds it very useful for their own personal reason.
• most used : I think Ansys *.msh format is most used in industry, but I don't know if other mesh formats are also very important to specific industries. Maybe structural people use Nastran's meshes.
• performance : mesh formats you like because of performance. I only use ASCII meshes till now, so I don't know much about this.

Thanks

PS : Cool project with many mesh converters : https://github.com/victorsndvg/FEconv
Mesh formats supported by them : https://victorsndvg.github.io/FEconv/formats.xhtml

[Simbelmynë]

Perhaps also have a look at the meshio project



https://pypi.org/project/meshio/

[arjun]

"Which grid formats should our solvers support for maximum support, user friendliness, and performance?"


There shall be one neutral format that shall be very easy to read or write by a simple C or C++ or Fortran code.

Then there shall be converters to it and from it. That would make everyone's life easy.

The project shall be easily available and easy to compile so that it is painless to use by any average joe.


PS: I created bmsh format for wildkatze based on this idea. Bmsh format is nothing but fluent file written without all those markers. Its basically, like (anyone can load this)

--------------------------------------------
Number of reagions

For each region

Number of nodes
x y z
.
.
x y z of nodes

Number of faces groups

For each group
for 1 to Nfaces
number of nodes in this face n1, n2, ... n_nn left cell and right cell.


-----------------------------------------

[aerosayan]

By "Number of faces groups", do you mean it supports partitioned meshes, or faces belonging to some boundaries like Inlet, outlet, wall, etc? I partitioned a GMSH mesh using METIS and tried to export it to Su2's format, but it didn't work.

[arjun]

Originally Posted by aerosayan

By "Number of faces groups", do you mean it supports partitioned meshes, or faces belonging to some boundaries like Inlet, outlet, wall, etc? I partitioned a GMSH mesh using METIS and tried to export it to Su2's format, but it didn't work.

No no.

I meant say for a pipe with 3 boundaries (all tetra mesh)

Number of face groups = 4


4

interior Interior
2 (number of cv faces)
3 n1 n2 n3 cl cr
3 n1 n2 n3 cl cr

inflow Inlet
2
3 n1 n2 n3 cl cr
3 n1 n2 n3 cl cr

out Outlet
3 n1 n2 n3 cl cr
3 n1 n2 n3 cl cr

sidewall Wall
3 n1 n2 n3 cl cr
3 n1 n2 n3 cl cr


simple to read and write

[arjun]

Bmsh is divided into two files

.info.bmsh

This contains only the mesh information like number of regions and their names and then their boundaries etc BUT no heavy data

.bmsh

Heavy data


--------------------

The advantage of this in Wildkatze is that you can do the whole set up without loading the heavy data. so if you have a billion cell mesh you can set up the case on laptop since the mesh won't be loaded at all.

Now once it is set up, you can put the whole thing on cluster and run the case.

[andy_]

Originally Posted by aerosayan

Which grid formats should our solvers support for maximum support, user friendliness, and performance?

The important format to get right is your own internal one. Then support whichever external ones your clients need and/or want. There are usually one or two widely used ones plus a few that are used less. It doesn't take long to write converters to/form you own format once you have done it a few times. Then when you find 3rd party converters are failing to function adequately (and you will) you have a viable way to fix things without having to trawl through some peculiar code written by a student learning to write computer software.