[David*]

Dear FOAMers,

I use an object of type Foam::meshSearch [1] to perform many (efficient) searches on a constant mesh in my program. However, it is based on an indexed octree, which takes some minutes to create for my mesh. Tree generation is currently performed on every start of my program, which wastes a lot of time. I need to modify and recompile my program quite often (doing research). I'm looking for a way to save time. My ideas are

1. Save and load the meshSearch object from disk:
- Use a serialization library like cereal [2] or boost::serialization [3]
- Use OpenFOAM ITstream [4]

However, serialization or writing seems to require a suitable new method in EVERY class/for EVERY non-standard type used to construct the meshSearch object, which are a lot (e.g. indexedOctree, treeDataFace, treeDataCell, treeBoundBox, diverse enums...). Most of them have private access to their data members, so I would have to create a complete redundant structure of C++ files in order to access them and build my own "meshSearchSerializable" object.

2. Create and access in memory:
- A server/client concept with shared memory using boost.interprocess [5]
The "server" would create the meshSearch object and continue to run, while the "client" (my program") would access it in shared memory. Could there be a bottleneck when like 100 instances of my program try to access the shared memory at once?

Could somebody comment on whether these approaches make sense or if there is an easier way, e.g. using other existing OF functions?

Thanks a lot,
David

---
[1] https://cpp.openfoam.org/v4/classFoa...eshSearch.html (I'm currently using OF4.1, but there is basically no change in this class towards OF9 or v2012)
[2] https://uscilab.github.io/cereal/
[3] https://www.boost.org/doc/libs/1_76_...doc/index.html
[4] https://l-infinity.de/posts/openfoam-oop-library/
[5] https://www.boost.org/doc/libs/1_77_...rprocess.intro

[s1291]

Hello,
Have you considered using PCL (Point Cloud Library)? or more specifically, pcl_octree module which has also functions for serialization and deserialization to encode the octree structure into a binary format [1].

References:

[1]: https://pointclouds.org/documentatio...p__octree.html

[David*]

This is interesting. However, if I understand correctly, it would solve my problem of storing an indexed octree, but would require some form of conversion of my OpenFOAM mesh to a point cloud. This seems to be more effort than my ideas because I also need the mesh to access field variables and interpolate them. Other opinions?



Thanks,
David