Is there a reference that shows how performance of the network is impacted based on the distribution across participating validators of memory, bandwidth, storage, computation, and staked amount? In particular, I am wondering which if any of those five inputs impose a constraint based on minimum across all validators on the shard vs those that impact performance based on the average or sum across participating validators. Some of the discussions make it sound like there is a minimum bond and a minimum hardware + bandwidth spec, and then the network performs at the level of the weakest participating validator. I have inferred, perhaps based on a false premise, that "upgrading" the network requires all the validators on the shard to upgrade their hardware + bandwidth. And it's unclear what the incentive mechanism / price signal is to coordinate that. Do the lagging validators just wind up getting slashed and eventually drop out if the rest of the validators upgrade?
There is no detailed analysis of performance as of yet. However, it is an item which the development team will begin to address in the very near future. Ideally, throughput will not be limited by the lowest spec in a shard and the average will be a more accurate measure. This would allow piece-wise updating of the network to provide incremental performance improvement.
These are all great questions. For the time being, the answer I can give you is that staked amount will not impact network performance. Your intuition that the weakest validator sets the network performance is close, I would say that it also depends on the level of clique safety one desires for a given transaction. If 20% of a shard's validators are weak, and I want a clique safety of over 80%, then I would be depending on some of those weak validators to validate blocks before I accept a transaction as final. But if I am willing to accept a lower clique safety, then I can proceed without necessarily needing the weak validators to contribute. Now, if a majority of the validators in a shard are weak, then yes, that shard is going perform more slowly than a shard that has high performing validators. If a validator proves to be too weak for a given shard, they will likely be ejected from the shard for non-performance. So when we say 'weak' validators, we have to bear in mind that the difference can't be so vast that the validator is unable to keep up with the necessary state transitions required for validation. I hope this answer helps.