I personally think @wpb has some very good points. I very much liked the sound of a standby subnet, and also very much appreciated @sat’s detailed breakdown of the challenges.
I wonder if two birds can be killed with one stone here. Another problem that exists is rigidity of the IC Target Topology. I believe it would be much more useful to have a target topology expressed as a set of tolerances (optimal configuration on one end of the tolerance, and the worst allowed on the other). Technically, if there are nodes sitting around doing nothing, they could be contributing to better decentralisation metrics for subnets (perhaps prioritising the most critical subnets, such as the NNS). Whenever a node is down on one subnet, an up node could be swapped out of an active subnet. This would reduce the number of nodes in the donor subnet, but that’s fine if it’s operating on the preferable end of the target topology tolerances.
Updating the tooling to make more efficient use of nodes would help the IC get more out of the nodes available. This will surely provide more options for mitigating growing pains as IC popularity continues to increase.
My understanding is that subnet membership mostly comes down to a registry entry. Do you really need to swap a node out of one subnet first before you can swap it into another?