Randomized Attention Test Design for Validator Monitoring in Optimistic Rollups

Transcript

from the network. Uh today I going to introduce uh the monitoring details in optimist rollup. It's called the randomized attention test shortly. You can call it rat and I'm really excited to give a talk in uh community conference and thanks for taking your time for attending. So uh I guess you are uh familiar with the rollups. So but in optive rollups the correct correctness L2 is assumed by deferred and it proven indirectly by fraud proof and uh it has dispute window usually seven days and during seven days

it guarantees that validators can challenge the incorre potentially incorrect the the state uh state by proposers. So the problem here problem is if the block proposers are submitting correct statement constantly then it means that you don't see any the onchain validator activities. So you know in in the real world there were only few the fruit proof cases. So it means in years of the optim ruleups um validators are very silent on chain.

So there's a concept called verifiers dilemma. Uh it was this concept was originally proposed in the by the loyu in 2018 80 uh 50 15. Okay. So uh shortly it means if all the other you are a verifier and but all the others verify then you are you are economically not motivated to verify because all the others verify the states then you don't need to right so however is it a real world risk it's a I think it's controvers but if valid but I think if validators are really silent uh I I don't think we can fully trust the

validators And also I think uh for for example in uh big rollups when uh big big defi apps they um deploy their apps in one specific rollup they try to uh operate the their own validator to mitigate the risk. So it means um still uh there's no big trust on validators and you don't see even validators are online or offline. So the this random the attention test is to propose to solve this problem.

So uh uh let's pass this slide. So attention test is a mechanism to verify validators are actually monitoring the chain. So to test their online and offline or they can they're following the state change or not. So this idea has been uh proposed and discussed for years ago. But uh the the thing is that nobody tries to design or implement this concept. So um uh why let's guess actually we don't know the real reason but maybe I guess uh there's a still believe that current security models are enough for opting rollups because the currently I I as far as I know there's no big hack about the optimistic rollup

security and uh maybe there's an assumption that validators are trustful or some the one reality is that option rollups are very currently still very centralized. So maybe it means you can trust them. But if we don't use these assumptions then we need something like another mechanism like attention test. So here's the core concept of the red. So in attention test attention test is randomly triggered. it should be randomly triggered with the posting the L2 state and second the target validator is randomly chosen and the attention test is a puzzle that's submitting the

closest Merkel leaves uh to construct a state route so for example the proposer just don't proposer doesn't need to change their parameter they can they con constantly submit their the L2 state but it can be just posting the L2 state or can be attention test. If attention test is triggered then validator should submit closes Merkel leaves that can construct the L2 state. So uh so it's very compatible with current of rollup designs. But here's a problem. Uh in the previous talk somebody asked about the randomness. Yes onchain randomness is can be expensive complicated. But we in

this uh mechanism we use um weak randomness by onchain and we uh I will address this problem in the the later slide and we will say this weak randomness something like using a block cache will be enough and uh I should say one point out one limitation that it this puzzle um guarantees that they are following L2 state but it doesn't guarantee that uh it will proceed access the full the dispute the pipeline because of the we can refer the chroma case there was one real world dispute uh challenge in chroma the one opinion

rollup uh unfortunately it recently closed their mainet but uh they they their validator validators found the the wrongly constructed the L2 state but they found at the same time they found their the smart contract can't process their dispute the challenge challenge. So they have to upgrade their smart contract to uh solve that problem. So here's a randomized uh attention test flowchart. So here are here are the existing rollup procedure stage transition post commitment by prop post commitment by proposer and waiting for fruit proof. If there's a fruit proof this game came and then then uh after

seven days finalization. So if rat can be integrated simply by with PA probability the proposer will post the attention test that is same the L2 the merkel root but in this if if smart contract decide it's not just commitment it's just test then designated validator should submit the closest the Merkel lives that can construct the Merkel root So if validator is live then the validator will submit the onchain the solution to the smart contract then if it's not live then it will fail then

timeout and even even if it passes the solving the attention test then still it goes to proof because uh the proposer and this designated validator share the same the state state change view doesn't mean that this state change is correct. So it we it still goes to the dispute window and if state mismatch or the validators the submission solution doesn't match to the L2 state then it will go to the dispute game and try to solve it and here's the flow uh in the view of the smart contract the proposer the

submits attention test puzzle with the L2 block number and the state state root And then the the RAT contract the this determines it will be the attention tester or not. And if test is initiated it emits event and uh if target validator is live then then it will submit the solution and the smart contract will verify very simply by just quick the hash the computation. So if so we can uh absolutely deploy the red mechanism but the problem is the frequency of this test the PA and the what will be the penalty. So uh we

um we analyze using the game series uh game framework framework the the full analysis is in the some the the paper paper in the last slide but uh let's go simple. So uh there will there must there are multiple the next equilibrium but the equilibrium we want is the the strategic equilibrium we want to force for the validator and proposer will be this right the proposer action is always honest it's it post only always honest state transition and the validator uh is always online so we want this

equilibrium and then Uh the result of analysis says uh uh okay and the result of analysis says that this graph. So for example if we just for for example there are like a 10 validators and if we put the attention test penalty failure penalty as like a like a $500 then we only need the 0.0 zero and 28 percentage per epoxy. It means the we need only zero and four tests per day of attention test the is enough to force this ideal equilibrium

and it so it means just one percentage of attention test trigger will force the the ideal security equilibrium. Then the question is you know we use the weak randomness right that it means the powerful the power or smart or rational proposer malicious proposer will try not to trigger attention test when try to state fruit state right then it it's a rational uh strategy but uh in this payoff table the in the fruit state trans uh posting case the the attention test the penalty doesn't affect this the equilibrium

condition. So, so that's that's why I'm uh I'm saying that this we can use just weak onion randomness and it means we can construct this mechanism uh simply. Oh, so what is the expected cost to construct this mechanism in smart contract? So, uh here are gas estimates. So state transition a state commit preexisting state commitment cost about 90k gas and if we post attention test then it gives 160k gas. So but uh it's mostly caused by

attention test information storage and attention test solution uh transaction cost like a 27k but we don't uh we don't post the attention test trigger every time it's just uh less than 1% is enough as I mentioned so the this cost overhead is very manageable I can say then what's Next about the this rat mechanism. I think the our game series the analysis should be uh purely reviewed in in a like academic level and uh there should be uh implementation and network network level

testing that can be released and and then why we need such a additional mechanism for optimism rollup because current dispute window is too long like seven days and uh as far as I know at the my basel knowledge 7 days doesn't have any theoretical basis. It's just consensus but and also so we need shorter dispute window for the for our user experience and one so we need I I can say we need the dynamic the dispute window and shorter dispute but window but the problem is engineering this the engineering and calculating this the dispute window is

very challenging but I think we can use uh we can engineering comparatively. So uh let's say the we can say 7 days enough with any additional mechanism then if we add such mechanism mechanisms then we can make it shorter than based on the seven days consensus and also another uh appro for the optim rollups will be definitely uh hybriding GK fruit proof because the current the bisection game of the fruit proof mechanisms are are very complicated so the yeah That's what I want to say and this is the end of the talk and thanks for taking your time and any questions and uh also

here's a paper of the this um this mechanism. Thank you. Um here you go. Thanks for the presentation. Um so quick question. How do you prevent cibils from verifying attention tests? because here I see you reconstruct the Merkel proof but they couldn't just fetch the storage from RPC or from another node and how do you fix that? Um I don't fully understand your question. Can you So if I understand it correctly the attention test is fetching providing a leaf from the state the following a Merkel proof of state of the leaf but what prevents them from using an RPC to retrieve that proof and use another node and hence you

could have cibbles just pretending to be nodes and passing the attention test here. Uh yeah that can be an issue. Uh good point. But that means the validators can still uh query such the states states from other validators. It means then still they can follow the states I think then validator should uh depends on such RPC then that that's the validator's own risk. I think uh can you discuss it in coffee break?

Yeah, thank you. Um any more questions? Thank you so much, Suan Lee. Thank you.