Well … as you know the more “rational” the mesh the easier to relax (and way faster).
There’s not a generic quideline for that mesh … but if we stick to the thread: Imagine a grid contained in the canopy boundary (kinda creating a uv mesh out of a BrepFace or dividing a BrepFace [with holes] etc etc). Now the x/y module sizes are variable meaning that you can adjust them to suit our columns (rational I do hope meaning the same footprint etc etc). So you get a grid of quad modules (all the same) that either are present or absent (columns). Then you can subdive that into also equal quads and finally triangulate (my advise: avoid that) the mesh. That way K2 gets a “uniform” input mesh to play with and obvioulsy works way faster AND yields a result that you can use for doing a rational W truss or a tensegrity truss (very expensive) or a hybrid one or a “dome like no W truss” or … something similar. Going from the K2 relaxed mesh to something with structural ability (with or without depth) is the 1M dollar thingy … since we do all that stuff in order to do something in real-life (OK, forget concrete ancient things - we live in contemporary times, he he). For instance (no “finished” columns in this case: waaaaaaaaay better aesthetics):
Even if you are after vaults … the same story:
BTW: Even if you have a tensile membrane in mind this approach works very fast: you just go from a quad arrangement into a cyclic one at the base (or top) of a given “column” that pulls the membrane (that’s why is critical to separate clothed/naked edges etc etc). See this linear arrangement (no “columns”) where the mesh is derived form subdividing the very simple flat mesh shown:
BTW: You can divide a quide BrepFace (non planar) that outlines the envelope and do the inverse with regard the anchors (unary -Z).
The same logic could being applied if you want hexagons or triangular column footprints etc etc [not valid for membranes] Obviously there’s some issues with pentagons and the likes … but the core of the matter is the envelope: nobody could complain about the shape of the column.