Great topic, some of my thoughts on this as well. I personally think there are two slightly separate issues here:
- Node providers contributing to ICP inflation - this doesn’t really seem to be the case to a significant degree as that appears to be mainly driven by other factors from some of the statistics shown and maybe that is where one should look first.
- Overcapacity of nodes (i.e. 50% of nodes just “awaiting subnet”).
In regards to point 2, I think a certain degree of overcapacity is necessary, in order to allow for sufficient capacity to be present when ICP starts to scale more, because adding nodes is not a quick process. Even assuming there would always be potential node providers ready to jump and add nodes, the steps to become a node provider are not quick. Just getting through all the proposal steps would take around 3 weeks at best. One also has to negotiate with data centres, and then, crucially, procure the hardware. The type of node machines required are quite high spec and at the very best procurement lead times are around 8-10 weeks….if there are chip shortages and the like, this can also increase to 6 months. Then one also has to ship them to where they need to go - particularly for non-EU/US based node providers this adds complexity, as they often wouldn’t be able to procure locally. So it can be quite a lengthy process to add nodes. Therefore having a certain overcapacity of nodes in the network at any given time is probably prudent, does it have to be 50%? Probably not, but maybe 20-25%?
Right now the top 13 node providers (each having more than 40 node machines) account for 66% of all node machines. And 87% of these node machines are EU and US based. I think this is probably where I would start to take a look. How many of these are Gen 1 machines? I understand it’s normally a 4-year business model and the life of the node machines are assumed to be 4 years, so quite a few of these may come up to the 4-year mark soon? (I don’t think this information can be found on the dashboard). They should then be phased out and not renewed. And given such heavy EU/US concentration and only nascent emergence of nodes in other countries, it may make sense to not allow any more additional nodes in EU/US and focus on adding nodes in the rest of the world for decentralisation. Yes, adding nodes outside EU/US is often a bit more costly because bandwidth costs are significantly higher and requires a bit higher node provider compensation, but this is also changing as time passes. And one should probably think about setting a lower limit per data center and location for the maximum number of nodes, to ensure adequate decentralisation (maybe 20 nodes or so, right now there are some of these top node providers with more than 60 nodes all in one data center). Lastly, Dfinity themselves run more than 90 nodes total in EU/US, I guess they must all be Gen 1 nodes. I understand if they wanted to run a few nodes themselves for development purposes, but that should only be a few, so maybe these can be the first ones to be reduced/phased out, while allowing for some additional nodes to come online in the rest of the world to drive decentralisation.