Solving Vitalik's trilemma with zk-driven DA&Storage - Igor Gulamov | ZeroPool

uh thank you uh I am Igor from Zer pool and um I will tell you about solution of vitalic trma uh with zik driving data availability and storage so what is uh the vitalic trma or the blockchain trma we have three properties of blockchain this is scalability security and and decentralization and we can select only two of three and this is an issue because we need the solution for Mass adoption um for scalability uh we need to keep scalability and the centralization with security um to uh provide infrastructure for uh Solutions uh like uh uh web 2 to web 3 transformation um and very L large amount of transactions uh so uh for example we know about forecaster uh it is using uh a lot of uh uh storage uh to store likes tweets and so on but um if vitalic trma is solve it then it's possible not only uh to uh provide uh infrastructure for these Solutions but uh it's possible to make uh web3 Instagram or web3 Facebook or web3 YouTube and also uh uh solve issue of transaction scalability to support not only finance transactions but uh um uh all decentralized computations we needed uh here is uh historical uh approach uh uh to solve the trilemma the first one is multi-chain ecosystem um it was inspired by poot and Cosmos uh the the idea is to make uh uh core engine and make a lot of blockchains on top of it and provide uh um blockchain interoperability due to this the same core engine um and another par parallel solution is plasma uh plasma uh provided high output and uh uh very early data availability like things uh uh based on Game Theory um the next evolution of ideas proposed in plasma is optimistic rollup and also there are Z car rollups and zums uh the issue of rollups is that rollups uh don't scale the data rollups just scale the execution and uh the uh last uh solution to solve this issue is data availability layer like ethereum d Charing and uh agent layer uh Celestia aale uh but all current that availability Solutions uh provide only shortterm uh data availability during two weeks or months this idea is originated from uh plasma when uh people decided then two weeks are enough for all and you have asset somewhere you will observe uh uh this uh uh object and uh if you see some issues you can uh interact with smart contract on L1 and withdraw your assets or fix the issues or something else uh but uh here is an issue because uh uh all users will not store the data of uh the rollups where they have an assets uh they should trust uh uh of chain infrastructure who stores the data and uh uh one of the things we are researching is how it could be solved by decentralized way uh so the rollups are not enough because rollups scale only data or only transactions but don't scale the data so how it looks like um the data is FL up to L1 or data availability layer uh the roller publish uh transaction on chain but also the roller publish block uh to to call data or to data availability layer and this is the bottleneck uh because uh call data is very very expensive and data availability layer will store the data only during uh two two or four weeks and uh we cannot perform with this infrastructure recursive rollups but we need recursive rollups if we want rollup Centric scalability uh we should uh uh uh have possibility uh to deploy rollups inside rollups and so on and so on but uh until the data is floating up we cannot do it because all data will float up to L1 and here is the bottleneck uh let's take a look uh at storage Centric approach because uh if we should scale the storage uh let's take a look at Solutions who provide the storage uh this is file coin rwa and a storage the general approach of this solution is that we have copies of the data uh each copy uh um May uh is made uh unique uh with sealing procedure when we add to this copy uh special High entropy vector and uh next uh nodes can mine rewards on chain uh with uh this uh data um with uh random sampling uh when uh we have a random challenge uh on consensus layer and next uh uh the nodes provide um mer lookups of the data uh these lookups could be compressed into Z zero knowledge proofs and uh nodes can get rewards proving that they are storing the data uh but we know uh file coin and uh and uh don't provide uh on the consensus layer high soundness for the data this is just replacement for web two uh Cloud uh data storage rwa provide the soundness but this is permanent storage and uh it could be not so efficient uh because uh for example if our rollups produce a lot of data we can publish the snapshot of rollup State and then we can forget uh the data uh before and then we we will use the storage more efficient and uh the infrastructure will uh spend less uh resources and money for maintenance uh our wave is permanent storage so we just uh send the file to the network and uh pay fixed amount and keep the fingers crossed that the more low help us and uh uh the sum of all spending during all time will be less than the amount we have paid and each storage provide configurable storage um but but uh uh speaking about uh our problem uh we need uh uh solution uh compatible uh to uh rollups uh I mean uh uh we we we need uh Z zero knowledge proofs of all execution all uh could be happening uh on the consensus layer of data storage solution uh it is possible um so historically uh the possibility of this was shown by Mina um and in our research we uh um show that uh all uh storage specific scenes like ceiling and uh uh random sampling and so on uh could be uh made inside U Circuit of ziki snark and then we can provide the proof that uh uh the data is stored and available this is uh uh important for uh uh solving uh vial stma by story Centric approach and rollups and another important thing is that all the solutions are using replication uh replication mean that uh we just uh copy the data to a lot of instances and we hope that uh at least one of CH one of instances will be available and uh the data uh will be readable to users and network uh read Salamon codes are U uh more sound solution because we can uh mix uh all the data together uh blow up uh the data and split it to chunks and uh that mean that for example if we make uh 64 to uh 512 blow up uh then any 64 shards uh are available to recover the data and this approach is 10 times more efficient than uh just replication uh let me show how it's working for example we have uh the data blob with uh for numbers uh we can uh uh represent it as values of uh degree 3 polinomial in specific points uh next we can blow up the data and now we have eight uh points uh then if we forget uh some some points and uh uh any four points are enough to reconstruct the polinomial uh because it's this polinomial is degree three and that mean that we can recover the source data uh here is how R Lon codes uh are working we can uh make blow up and next we can split the data to the network and then uh if uh we can uh take take back uh enough uh data then we can reconstru the source data uh one of the issue of of Ron codes in Data Solutions is that each node should know that uh the chart is corresponding to uh initial data commitment and uh with uh this polinomial equation we can easily uh uh Pro prove it that uh uh the commitment is corresponding to the all uh so data uh we just represent the data table uh of numbers and uh uh calculate the uh this uh table as bant polinomial and then we can uh perform random linear combination of columns and prove that um this is correct random linear combination of columns then if we have enough random linear com com inations of columns uh on different instances we can recover the initial table uh so uh here is minimal uh Z Driving data Centric rollup uh we have here uh uh data storage network with mining nodes we have uh ink um uh smart contract uh on which uh the no can mine uh the data uh mine rewards with the data and uh then we can join the proofs uh that the data is stored and available with uh uh execution proofs of rollups and then we get uh ZK uh proof uh that uh uh State transition of rollup is correct and the block data is available and this is enough uh to to uh uh be sure that every scene user need to uh use this rollup uh is correct and available and we don't need then uh any additional uh data availability um uh Solutions uh so and the uh main issue here that uh uh uh uh uh the the main solution here is that rear cion is uh unlocked why because uh uh if we uh provide uh on rollup level uh data availability and execution uh proofs uh then we can deploy rollups inside rollups and uh no data floating up to L1 and we can uh uh deploy rollups inside rollups uh uh if we need a loot uh uh a lot of time that's we need um so here are some parameters of data Centric rollup uh we compute a storage cost for uh um this taking into account header node uh 15 cents per Gaby per year um and soundness is more than 110 bits what does it mean uh if half of network is malicious uh then uh the probability to lost any uh chunk of data is less than 2 in power minus 110 it's a very low number and capacity is 1 petabyte uh so uh next uh and roll uption is unlocked uh next could we reach more scalability uh the single uh rollup instance uh capacity is more than one petabyte but when Lo at rollup recushion it mean that we can uh build ecosystem uh based on recursive rollups uh it's a two- layer solution when uh on the top layer uh rollup operate which uh mining nodes should uh store which data uh this is important because uh if we have no such layer then uh malicious operators can concentrate to single segment of the network uh uh we uh make uh modeling that uh when one uh uh note is going offline we need just make uh two random swaps between all nodes and this is enough to provide High soundness uh next uh we have uh on the second layer these rollups I described before uh with more than one PAB capacity um why uh it is important uh because uh with uh uh scalability uh of data uh we can uh make web 2 to web 3 uh transition um so I as I told before about foraster it used its own data availability solution with uh also random something but with replication uh uh using uh decentralized uh uh high sound solution based on Ron codes we can provide infrastructure for uh uh Solutions uh like decentralized uber uh decentralized Facebook decentralized Instagram and maybe decentralized YouTube also uh and uh of course it is possible to make uh uh High soundness data storage for existing rollups uh when the the data could be stored uh um enough time for all users and uh uh the data could be removed only after the snapshot of rollup is published so when it is outdated and doesn't need it to uh make uh transactions uh here are links for our research related to this solution uh thank you for your attention if you have questions uh I just yeah thank thank you for the talk so uh am I correct that to achieve like full decentralization we not only need the proof that the data is stored somewhere but also that the data is delivered in time to some node so say if I want to become an independent independent sequencer or independent prover or something like that I need to have the data itself so could bandwidth become the bottleneck then or what's you your thoughts on this uh uh that's good question uh we U on this layer with Z snars we can prove only that uh if uh uh 15% 50 50% nodes are honest then with high probability uh the data could be recovered but uh um to make last uh uh content delivery system we can uh perform additional layer uh so uh uh uh this is uh another uh another problem uh it could be solved also with more classical approach uh because uh this is very relative thing I mean uh if you on different uh uh parts in the world in in Australia or in in uh India or in Europe or in America uh the uh uh throw putut will be different I think this is the problem how to make contain delivery system [Music] uh so I have a more question from the implementation perspective so let's say we have a ZK rollup uh which is the L2 to to ethereum and my question is since you propos that you are not making tra tradeoffs to trama is it possible let's say to build rollup that will use your da and uh how it will affect the throughput of of rollup if we do not make a trade do to trma so that's the question so basically the question is if we can uh go beyond the TPS ceing of current PRS uh uh thank you for your question uh so um um our solution uh solve U uh it it is not uh uh we we cannot get all three properties of course it's tradeoff anyway all solutions of the TR is tradeoff uh so uh What uh the tradeoff means in our solution for example we cannot build Unis swap with uh trillions or quadrillions of transactions per second uh between all uh uh accounts uh in this system this we have anyway multiple segments and uh ping between the segments is uh uh longer than inside so um we we can U uh take existing rollup and we can deploy multiple of these rollups together and all of these rollups can uh publish the data to this data availability and storage layer and the rollups can uh use some message protocol uh to make transactions between them uh this protocol will be slower because uh uh it need upper layer uh to make the transactions um um I I I think the solution of parallel evm execution um is solved Now by another teams uh our solution is data Centric uh so thank you for your question any more questions when you talk about transition from Web Two to web 3 how you're going to evaluate the significance of keeping data because of course maybe some data are just computer logs maybe they are crucial for some point of view but but maybe not so diff not so uh strict requirements than for example financial data especially when we talk about banking or maybe some personal data of the citizen and so on and vitalic also in his last article said that he's going to create like a scale from the finan on one kind of the range and the data it's like the latest tweet on the another because we can put data in different graduation and some of them should be really have a very strict protection maybe it should be backup on a multiple devices across all over the world and some could be considered no strong it could also make a less so strict requirements for infrastructure if we would consider any piece of data as something very very significant we can really end up these resources anyhow how you're going to solve the problem in your solution uh thank you for your question very nice question in our architecture the soundness is very cheap what does it mean that mean that if the data blow up is eight uh then we have very high soundness it's okay for financial uh tools but if we have for example soundness 4 or two uh then soundness is low uh so that mean that for our solution it's better to uh provide on consensus layer soundness more than 110 bits for everything uh thank you uh [Applause]