Benchmarks

[fdrmrc]

• Geometry: piston
• Number of elements original hexahedra: 365 712

The time considered here is the total time required to partition the original triangulation. Therefore, with the R-tree I sum the following steps:

1. Build the R-tree
2. Visit the nodes of a specific level and store the information about the agglomerates
3. Effectively agglomerate elements of the original triangulation together

With METIS, I record the time needed by the deal.II partitioner function. This function internally calls METIS_PartGraphKway() and flags the cells together.

---

• Geometry: brain (3D)
• Number of elements original hexahedra: 1 523 408

[fdrmrc]

Another view of some agglomerates inside the piston mesh:

[andreacangiani]

Very cool!! You say another view...do you have other nice pictures like this one?

[fdrmrc]

Sure, here they are (let me know if you prefer other views):

[fdrmrc]

Btw, within the next few days I plan to compute the metrics related to the quality of the agglomerates (taken from the "Neural Network paper").

[andreacangiani]

Fantastic, thanks! The timings above are striking!