Provers: To decentralize or not to decentralize?

🔬 Research
1

Proposer (sequencer) decentralization is all the talk now. But what about provers? Should rollups have decentralized provers?

Here are some of the for and against arguments that I’ve come up with or found on the internet.

For prover decentralization:

  1. If you have a centralized prover (like all ZK-rollups now have) and it goes down, who picks up the work? One solution is to have some sort of a pool of backup provers who could step in. In this case, the DAO or any other governing body that owns the protocol should pass an emergency vote to open the gate for other provers. However, this would introduce some delay.

  2. If there’s one centralized prover, all prover rewards go to its pocket. This could be mitigated by some type of a mechanism that would redistribute the prover rewards back to tokenholders. But this might mean that proving is an unprofitable activity for the rollup team that’s running the prover.

  3. Even the small fish could join the prover network by using resource-pooling services. I believe this would work similarly to liquid staking: You add what computational resources you have to a large pool. The pool then uses those resources to run a prover and generate validity proofs. The rewards that the prover receives are shared between those that supplied their computational resources to the pool.

On top of that, proof computation costs and requirements are expected to go down with time, allowing for a larger diversity of provers.

ZKPool seems to be building something similar but there’s not that much information on it yet. Check also Ingonyama.

Against prover decentralization:

  1. Decentralized provers are unable to censor transactions and don’t increase security. The worst they can do is delay onchain finality of L2 batches.

  2. Decentralizing provers would add unnecessary complexity to the whole thing.

  3. Multiple provers can be working in parallel to generate a validity proof for a single block but only one can succeed. This means that other provers who tried to generate a proof for the same block have wasted their resources. Also, there’s the opportunity cost.

Also, some additional things to consider:

  1. What do “decentralized” and “permissionless” mean? Are they the same thing? Can a network be decentralized but not permissionless and vice versa?

These are just thoughts from my simple mind. Galaxy brains, come and show us how it’s done.

Some interesting proposals/discussions:

  1. ZKP’s Two-Step Submission Algorithm: An Implementation of Decentralized Provers.

  2. Based rollups—superpowers from L1 sequencing - #14 by fradamt - Layer 2 - Ethereum Research.

19 Likes
2

I think this topic is very interesting.
In fact, in Taiko Alpha-2, there were 194 pre-selected Provers, and the Reward ($TKO) share was not fair.

However, in order for Taiko to achieve true decentralization, there needs to be a mechanism to prevent monopolies, but this would add the complexity you speak of.

This topic seems to be something that needs to be worked on. I’m glad we’ve had a healthy discussion since the forums came into existence. thxs.

8 Likes
3

Hi, the provers in alpha-2 were not pre-selected. Anyone could’ve participated in the prover network, it was permissionless.

However, you are right that only the most powerful provers can now expect to generate proofs. This is something all current ZK-EVMs would have if they decentralized their provers. But as I said in the original post, one way to solve this would be using ZK resource-pooling. I haven’t seen any explanations on how exactly this would work though.

3 Likes
4

It has almost been a year since this topic has been brought up and I was curious what developments or ideas has been made/proposed in case of “fair” way of proof incentives.

I couldn’t really understand what the main goal or main problem Taiko wants to address. Daniel said in his interview with RJ that he wants to create a tool or mechanism where even the poorest can have access to financial tools at the lowest cost. Is that still the case?
If so, how would “fair” incentivizing proofs be fair if we need to keep the cost down which actually means that proofers need to fight eachother and be real competative. How does this motivate proofers to help this poor guy to get the financial help he needs?

Does it make sence what I am saying? Maybe I am just delusional I dont know :slight_smile:

5

Interesting read. I would love if someone were to expand on the suggested proposals/discussions.