There are more churns in simulation 2 simply because the network is bigger, and it is bigger because there are less nodes that cannot connect. So, no downside in this area.
The number of churns is even smaller if we look at the figures when simulation 2 had a similar number of nodes:
| Metrics | Simulation 1 at step 890000 | Simulation 2 at step 260000 |
|---|---|---|
| Adds | 800814 | 234114 |
| Drops | 62523 | 18053 |
| Rejoins | 26663 | 7833 |
| Relocations | 503815 | 376067 |
| Rejections | 578686 | 56978 |
| Churns | 1912122 | 1021403 |
| Sections | 8258 | 8358 |
| Section nodes | 161447 | 163394 |
| Left nodes | 26287 | 3997 |
| Rejection rate | 72% | 27% |
Simulation 2 is much better than simulation 1, except in the number of distinct prefix lengths which varies from 3 to 4: the values are better (I mean prefix lengths are greater) but they are slightly more dispersed than in simulation 1.
So, I improved the program to make sure that when a node is relocated to a section it is placed in the section half having less adults. The simulation is running right now. Results are expected in a few hours.