Do you have any idea when this happened? It’s slow in my 3.4.9, so the change must have occurred quite long ago.
When I use the Deep Connections Gramplet, I can also see that it thinks that a spouse is more important than an ancestor, which in my case is quite silly, because the route through that spouse does NOT result in a shorter path.
The other sillyness is indeed following siblings, and passing through the same person twice. And a good algorithm can work very well without artificial limits like the number of generations that we now need to set in preferences. Even 20 years old PAF doesn’t need that, nor any of our current competitors.
It could be that my tree outgrew what Deep Connections could handle efficiently. (So performance didn’t degrade on a curve but dropped off a cliff.) But no, I am unaware when performance degraded. My use of the Gramplet was too intermittent.
OK, in that case it’s likely that your tree outgrew it. The component doesn’t scale very well, and it is very sensitive to changes in the generation limit.
I often use RootsMagic to exchange data with FamilySearch, and that program uses a fully normalized database on SQLite. And when I run that, in Virtualbox, with only 4 GB RAM, and 4 of 8 cores allocated to the VM, on my i7 running Linux on a HDD, it does many things faster than Gramps, like GEDCOM imports, and searching for duplicate persons.
This suggest that, althoug writing data to several tables in a full relational database is slower than writing a single row with a blob in it, the total performance can be better. And I guess that’s because the database interface in RootsMagic is simpler, and faster.