My imperfect understanding is that zkrollups do function ok for simple payment-style transactions. However once smart contract computation gets involved it becomes very difficult to generate the 'proofs' in a cost-efficient manner.
These "prover pods" (ie. a centralized AWS group of servers) can be trivially DOSed by people submitting insanely costly smart-contracts that will take down the entire pod (ie. smart contract that ask you to factor huge random number primes or more amusingly a smart contract implementation of BTC PoW).
Lastly security is an issue as how do other prover pods verify the proof results. Its not totally clear the exact way that would happen....So if a very expensive proof is submitted (suppose its not a DOS attempt, just an expensive computation), and a proof is generated. How exactly are these proofs verified....in many cases like prime-factoring, hash generation, etc it may be trivial to verify that hash(xyz)=abc...but depending on the algorithm re-running the proof may be just as costly the second time as the first...so you can see how computing time could spiral out of control.
Right now if you watch discussion on these issues, the ETH crowd mainly does a fair amount of hand-waving on these issues and provides little in the way of concrete solutions.
off-chain labs did a write up of Optimistic rollups vs zk rollups and they explain why zk rollups may also be bad. So basically Optimistic rollups beat Zk rollups but sidechains beat optimistic rollups.
reply
In the end, Vitalik will probably just re-brand LN by changing some terms and altcoiner grifters will just pretend like they invented something novel.
reply