Performance Based Node Rewards

Not at all!

Not necessarily. We’re basically hoping or counting that the spatial autocorrelation / Moran’s I will hold, where nodes in the same DC should behave the same way. But I agree - it’s certainly not something that is absolutely true in all cases. For instance, if someone buys nodes in several batches, they will likely have different failure rates. However, most people will buy nodes for the same DC from the same supplier, and in a single batch. And the nodes would share the same power supply, cooling, internet uplink, etc. So I would expect the nodes in a single DC (from a single operator) to have similar reliability. But you’re right, it’s not a strong proof.

However, if we would add some nodes into an idle subnet, with almost no compute/storage/network requirements, that’s a very bad indicator of how the same node would behave if added into a very busy subnet with heavy compute&storage&network pressure.

I’m not following that question, sorry. Extrapolation wouldn’t work if someone has 1-2 nodes, as this NP currently has. I don’t see how this (spatial autocorrelation) could be done if there is nothing to correlate to. We could do temporal autocorrelation, possibly, by assuming that the nodes would behave in the future similar to how they performed in the past. But that has its downsides as well.

Note that this particular problem we’ve seen in here wouldn’t be solved if we used the “big evaluation subnet” approach, since the node experienced a hardware issue and this could happen at any time regardless of whether a node is in a subnet or not. So I’d rather concentrate on the cases like Seoul vs HongKong DC, both managed by the same NP and behaving drastically different.

1 Like