Modular March EP 18: Unveiling Blobstream and Celestia's Architecture, with co-founder John Adler

Transcript

[Music] [Applause] [Music] he welcome back to another episode of the rollup and modular March modular April but we're keeping it going uh today we have the pleasure of speaking with John Adler uh from Celestia a really core

piece of ethereum scaling um one of the original Believers of the rollup Centric road map and of course Rob welcome in John it's great to have you in and um excited to chat about the recent developments glad to be here thanks for having me well uh let's start by uh unpacking Celestia um Celestia has a architecture of a various various types of nodes all performing uh their own function you could even call these nodes Specialized or modular in their own right um and they band together to form the uh framework for modular blockchains making

it very easy for developers to deploy their own modular blockchains um each of these notes have very specific use case function in the Celestia network uh so John could you start off by telling us about the nodes in the Celestia Network yeah so I actually tweeted recently about this uh but there's several there's several different types of nodes so if you you know hear someone on Twitter say oh St nodes need to they need to be a they need to do a certain amount of work or whatever then you should automatically there should automatically be be a red flab because there's no such thing as a celestian node uh that does a specific thing uh likewise there's no such thing as an

ethereum node that you know does a specific thing rather there's different types of nodes and each and each type of node has different properties uh security properties and also different uh different things that it will do uh this applies for every blockchain you know from Bitcoin totia to ethereum to salana well I don't know about the last one but let's just say at least for the Bitcoin and celestian ethereum uh so what are some different types of celestian nodes uh and why uh and why is it that some of these nodes don't necessarily exist in a normal blockchain but they do exist on Celestia and that's and the fact that they can exist on Celestia is one of the reasons that

Celestia can have some of thece properties that it does have so first is the Celestia fold note so this one here is going to fully download uh all the block data on Celestia to verify that it is completely correct and also to verify that it can reconstruct the blocks uh this is necessary because even in a blockchain that has data availability sampling which allows light nodes to to determine whether or not a block is being withheld there's still kind of this corner case where uh if a block cannot be reconstructed then uh then like as a whole network the

block is not available it's kind of a corner case in the data availability sampling protocol and this exists regardless of the cryptographic specifics of how you implement or like the underlying cryptographic specifics of your data availability samping regardless of you use fraud proofs or kg proofs or ZK proofs or whatever there's still this corner case this Edge case where uh you may need to reconstruct the block uh so full not uh perform that service uh they will fully download blocks and attempt to fully reconstruct them uh obviously they're more expensive to run so we would expect you know a fewer number of them uh but one thing that's very important is that full nodes aren't

necess aren't the aren't the aren't required to or it aren't required for node Runners to contribute security to the network uh which I'll talk about in a second so on the other end of the spectrum of full nodes are light nodes so these light nodes perform data availability sampling rather than fully downloading blocks so this means uh they will check uh very small portion of the block of every block uh and importantly it's not a p it's not a portion of the block that is proportional to the block size rather it's a fixed number of parts of the block regardless of big the block

is the block could be you know one gab big it doesn't matter you're still only checking the same number of different parts of the block and the parts all have the same size uh and through this process of data availability sampling combined with enough light nodes performing this data availability sampling the network as a whole can uh can determine securely whether or not block is available or not uh without without these light noes having to fully download blocks so this is uh this is kind of an amazing technique that allows Celestia to scale while also allowing users to verify that the Celestia blockchain is correct without having to do a lot of work uh because like we know one of the reasons that blockchains have any sort of limit

on their throughput is to allow end user verifiability right if we don't care about end user verifiability then just you know run it in some web to AWS database uh the whole reason that blockchains are so important uh uh like the the one of the very important properties is uh the ability for end us end users to verify that the protocol rules are being followed by the block producers okay so uh now these light nodes uh don't do reconstruction I said there was I said there's an edge case where okay they can perform sampling but they don't do reconstructions so who does reconstruction uh we could say well full noes do reconstruction of the whole

block that is that's correct but can we have uh can we have reconstruction performed by a node that is not as expensive to run as a full node and no that does not fully download blocks it turns out you can uh you can have uh you can have nodes that don't have that aren't that don't require fully downloading every single block and these node can still contribute to the security of the netw how does this work so there's kind of two orthogonal axes that uh what we call partial nodes U but that could also these could also be called sharded nodes there's kind of two orthogonal axes on how these partial nodes will split up their

responsibilities one of them is they can download parts of every block the other one is they could download uh not every block but every whatever number of blocks I say one every thousand blocks uh then uh these partial nodes can perform the work of reconstruction can perform the work of checking that the protocol rules are being followed uh and this act of checking that the blocks are correct the nice thing about Celestia is that this check can be performed stateless in other words it does not require nodes to know what happened in every previous Block in order to check that the current block has been constructed correctly because the

construction like the the Reconstruction process that I was talking about is completely stateless it happens on a per block basis it doesn't happen like it's not something that requires you to know all the previous history to then verify the current block since it's stateless it means that you can actually Shard this work you can distribute it across multiple different nodes and have each of them be responsible for a subset of the all the blocks for instance if you have a large number of partial nodes each partial node can instead of downloading every instead of fully downloading every block they could fully download let's say one out of every thousand blocks uh and in this way they can perform this reconstruction process if needed on those blocks uh without having to fully download every block which and

synthesis St this process it means they're contributing security to the network by there existing some nodes that can verify this Edge case verify that the rules of the protocol are being followed correctly by the block producers uh and these paral NOS can be very cheap because they don't need to download every block they can download every you know one every thousand block so now you can run it on a very cheap computer computer versus requiring a very expensive computer or a powerful server that would have to download every single block and again this is a complished because the building each block is a stess process and in a traditional blockchain such as let's say ethereum Bitcoin salana the vast majority of the work that's being done

in validating the block is not is not uh independent of other blocks but rather it's about processing transactions um uh where the effect of a transaction depends on all the previous transactions in the blockchain so you need to know the state of the blockchain you need to know everyone's balances to check if people have enough balances to know the current you know you need to know the current amounts in the pools if people are doing a swap uh etc etc you need to know you need to know all that and to know all that you need to know everything that happened in the past plus the different is that you can do things on a per block basis uh so this is uh part of kind of what I was trying to elucidate in my short Twitter Thread about this topic is that Celestia partial nodes which you could also call

sharded nodes uh allow users to in the like the worst case cost that a user needs to pay in order to contribute security to network is actually very low if they run a partial node and they don't need to run a full node to contribute Securities to the network full node just do extra work which is which is like a nice to have kind of like on ethereum ethereum has these archive nodes that serve historical St and it's very convenient for the network to have archive nodes but if all of the archive nodes suddenly disappear the network wouldn't just become insecure no because you can reconstruct an archive node from a full node you just need to do you just need to do more work so it's a convenience thing not a security thing similarly the

existence of Celestia full nodes is a convenience thing that it could be more convenient for someone to connect to a single full node that has all this data versus multiple partial nodes but it's not necessary for security for users to run a full node to contribute uh like full nodes do not contribute any unique security properties to the network they're purely a convenience thing okay so I hope that's kind of a summary of some of a subset of the different celestian not types there are s Al there there validator nodes there archive nodes Etc but these kind of form the like fundament there's fundamentally you know a few different celes no types but these partial nodes or shed nodes if you want to call them that are are something that all the blockchains that don't

operate as a data availability Network ones that aren't stateless in the way the Celestia is just cannot have just not a thing that's possible in those in those NWS and it's something unique just SLA so John I want to hone in on the word of statelessness because that's something that I've heard some other protocols use but I just want to make sure that we're all on on the same page when we say statelessness so the the idea that we unbundle historical State and current state means that we can put the historical state in the archive nodes and we can we can verify the current state using this sharded node or a light and I kind of want you to help

me here because we we kind of push all the historical State into the archive node and then it means that these partial nodes are able to operate in this stateless way is and and because they're able to do that they're much lighter and it means that you can run them with much lower no requirements such as running them on on your own CPU or uh rather uh like low storage or low low performance type of uh GPU or chip is that is that a a fair assessment of the difference between these archive nodes that store historical State and then these uh partial nodes that are verifying the current state so the the assessment and the

conclusion are direct correct uh there's some some there's some wording uh there some wording things that aren't exactly precise so I'm going to pull on my you know I'm I'm going to be really pedantic here and and and kind of talk about the distinction between block data and State uh and the lots of people get mixed up between the two because they have a lot of similarities and in normal blockchains you kind of need to keep both around uh but so the difference is that state is the state is whatever is required to verify new blocks it's something your node your full node or your I guess partial node as it were uh

needs to keep around in order to verify the next block uh block data is things that are only needed for the current block and then can be dumped into some you know something like spinning a disc something that's cold stores that isn't accessed very often doesn't requ it doesn't require Random Access basically uh so those are kind of the two different things so in a normal blockchain like uh ethereum for instance that do smart contracts that do execution Etc uh the state can be quite large it can be hundreds of gigabytes because that contains all of the account balances Etc uh and blocks are actually quite small uh and blocks are largely transaction data that mutate the state uh but on a

blockchain like cestia it's quite different uh in the sense that what is really need what is needed to be kept around in order to validate a new Celestial block if you're running a full node or a partial node uh this is where I kind of skimmed over this detail the paral nodes don't keep around no State they keep around the celestial State uh but then they don't need to fully download all the block data and it turns out because the Celestia State machine is very minimized it's very minimal uh the state they need to keep around is very very small we're talking on the order of like 100 megabytes or something uh it's very small so it's not zero so if I if I call like partial no stateless it's an approximation in the sense that compared to ethereum they're essentially

stateless technically they're not uh but the distinction is basically that uh these partial nodes don't uh these partial nodes don't have to do a lot of work they do do very very minimal work to actually process State changes but a large a lot of the work that they're doing 9 99% uh is going to be processing block data and plot for block data you don't need to keep that around to process the next block and the next block you only need it to process the one block that you're processing State needs to be kept around to process all future blocks uh so that's kind of the distinction in the sense that the vast majority of the celestial block work

that's being done is processing block data not State transitions the vast majority of work in traditional blockchains is processing State transitions which have an effect on the state and all of that needs to be kept around you can't put that in to call storage and very little is about processing of the actual block data does that does that answer a question it does it does I got one follow-up question and I'm sure Andy's got got questions as well my my follow-up question is this idea of reconstruction I want to I want to so I want to try to unpack the difference between reconstruction and verification um so let's say I have let's say I know the state at T2 do I also need to know the state at t0 and T1

to make block at T3 or do I do can I just use T2 and all and then reconstruct uh and then verify the block that comes next at T3 so this is where the distinction between the state and the block data comes in so you don't need historical States so you don't need to know what the old state was you only need the latest date and and in general in any blockchain you only need the latest date uh but the question is also around the block data is do you need any block data and the answer is you don't need any block data on any blockchain celesty or not uh to process subsequent blocks you only need the state so in Celestia the state is almost non-existent so essentially it's stateless to process new blocks compared

to traditional blockchains uh while in a traditional blockchain since you 99% of the work it has to do with State uh then you need you need to keep all this state around to process new blocks does that answer a question yeah so in a traditional blockchain you would need to know the the previous blocks in order to uh construct the new block but not not in you need you need to you need to have processed the previous blocks to get the latest date uh like you would need to process the previous blocks you would need to process all of their state transactions to get the latest date while in Celestia since the majority of the work is block

data you would not need to process the previous blocks block data to verify the current blocks block data and that's that's the important distinction that's why it's so powerful to have a blockchain that's you know nearly stateless that the 99% of the work is block data because that can be sharded you can Shard that responsibility since you don't need since a node does not need to download every single block to process new blocks they only need to they only need to download the block as it as it comes in whichever whichever when they're assigned to or whichever one they want to verify which could be you know a very small fraction of the total number of blocks and that means that you can verify one of the or you can download one of the blocks I can download a

different block and because I don't need to have downloaded all of the blocks I can just I can just download mine you can just download yours and then we can reconstruct the new the newest block together yes cool okay you can reconstruct it together and you can also reconstruct it independently okay okay I think the big distinction is like there's no smart contracts on Celestia the only thing that is being H that only smart contract execution is staking I think and sending sing and transfers uh there's a few other very very minor things but

basically yeah like there's no like Unis swap on Celestia right there's just no capability for that and it was purpose built for that so exact we we can then create this idea of like okay this blockch and that's where the whole Moder thing comes in but to kind of rewind so we've got full nodes which are are are are the the highest cost of coordination for uh reconstruction or a reor which is effectively a catastrophic event and you've got light nodes which are like able to uh facilitate data availability sampling um which is almost like the the magic sauce of Celestia is is this idea of Das by light nodes then you got these partial nodes which uh we recently had a live stream with the founder of Igan lay and we talked a lot about the cost of

coordination and how having to go back and download the entire full state of a blockchain um is quite is quite Hefty and and the and the eth classic Fork was just this big example of like how much that costs the ethereum network but we've got these partial nodes where look we we don't have to go and rely on the full nodes uh to kind of uh give us that that that state back we still have this kind of lightweight version um and you keep bringing up Bitcoin ethereum and salana and and something tells me that that you've seen this chart that was circulating Twitter this weekend which is a a chart with a yaxis here of um small blocks and large blocks and and and uh non VM and VM and we've got Bitcoin in non VM and small blocks we've

got Celestia in large blocks in non VM and then ethereum in VM small blocks and salana in VM Big Blocks um and you you've been kind of touching all all on this and honestly I just want to understand what you you think about this this kind of framing I actually hadn't seen that diagram or the chart whatever you want to call it that uh but I can imagine in my head I guess I guess it's I guess it's a decent ging yeah I can see I guess it's a decent Fring I'm not sure how you useful it is in the sense that like it's true but

also how useful how useful is it to like you know Bitcoin can Bitcoin act as a modular data layer just because it's a non VM the answer is not not really also Bitcoin does have a VM just because they don't have just because Bitcoin doesn't have like a Turnin complete VM doesn't mean they don't have cont they do have Smart contracts there with Bitcoin script yeah but I think this is this is almost what you you were getting at is like okay Celestia has these large blocks and it's there's no virtual machine there's no smart contract logic there's no execution very minimal perhaps it's stateless um and so it kind of fits in this in this new quadrant like there's not a there's not a quadrant here that

is uh um kind of you know part of this in in the past so I'm just going to pull this down and we can continue the conversation so we understand the celestial architecture we understand the the nodes we understand how data availability sampling works and also why Celestia has been kind of made for uh data availability and for data publishing so now we've got blob stream coming in and we've got this idea of data attestations we've got uh ZK proofs with succinct uh blob stream has just uh Gone live in testnet for AR arbitrum orbit and Bas um yeah what is blob stream for somebody who has the a

fundamental understanding of data availability but not a not an extremely technical one sure so uh blob stream kind of give the tldr is uh a oneway relay I'm going to avoid using the term bridge and I'll say why in a second it's a oneway relay of uh attestations around the fact that data is available on Celestia to another blockchain in this case an e m one or more evm compatible blockchains such as ethereum arbitrum base optimism whatever whatever else is ethereum compatible uh in the future I'm sure we can have

additional additional blockchains to implement a blockchain verifier uh you know such as something in this n ofm why not or with bit PM why not uh there's nothing inherently stopping us uh from from doing that so essentially allows it allows all the blockchain to uh to leverage Celestia for data availability uh so you can essentially use Celestia and an ecosystem uh and how does it work and how is it different than a bridge a traditional Bridge uh so a traditional to token bridge is two-way and it also involves the movement of assets tokens uh which has a few downsides one of which is that

if you manage to move tokens across the bridge without some sort of authorization it becomes very hard to CLA them back as soon as anything happens with those tokens right as soon as they go into like some sort of decks as soon as they go into a mixer then it's like it's it's over you have to like you do you do a roll back right uh it's basically it's it's extremely difficult to kind a roll that back uh because it's it involves tokens uh and is also two-way uh which involves you know additional communication complexity Etc uh uh blob stream is different in both of those ways in the sense that it's unidirectional it's one way it's an attestation around the fact that some data is available in Celestia and this

at station is related in one direction from Celestia to another blockchain let's let's say some let's say some evm compatible blockchain just for the sake of Simplicity because that's uh there is an uh there's a verifier contract written for the evm that exists today not so much for other other environments uh so uh what do this give us over other schemes for leveraging offchain data availability Solutions such as data availability committees so one of the issues with data availability committees and all the protocols uh such as restake uh data availability Solutions or anything that's entirely on chain is how do you do

slashing the the one question you should ask whenever someone says okay we have some mechanism to increase the data availability through puts through some offchain mechanism for your evm based chain you should ask okay what are the security guarantees well you have to ask two things one of them is how much is bonded so how much can be penalized if these actors misbehave and the second is how do you P lifestyle and you need both and a lot of people will only talk about the former they'll say oh our protocol has so much money bonded and it is you know all half of ethereum stakers are bonded in the reaking protocol and there's so much security but if you can't penalize against this Bond uh then

it's it's worthless it's not the bond means nothing right a bond is only is a bond is only as like as is important it's only as useful as you know how much you can actually penalize against a bond if you can't penalize against at all against it at all then it's just a deposit that can be withdrawn it's just Capital inefficiency so need to consider both cases uh with uh with the data availability committee there's no bond it's purely just reputation based right you have whatever five seven whatever actors they're highly trusted they're highly wellknown if they misbehave uh the reputation is in the uh kind of like uh let's say I don't know let's say Silicon

Valley Bank right uh let's let's use that one how about uh in the absolute man absolutely I mean the reputation is pretty bad but you know has that helped their customers not so much right they needed some sort of centralized bailout uh so reputation doesn't exactly doesn't actually doesn't it's not exactly worth much especially you know once the thing's already collapsing then whatever it's it's done uh the other alternative is you have uh you know some sort of onchain protocol some sort of committee that's onchain uh you know it could be what what whoever however bought themselves right it's some something on chain like a and is ultimately

some isomorphism on a doubt the problem with this is how do you slash how do you actually determine since everything is purely onchain how do you determine that uh these actors have misbehaved offchain uh data availability is not a uniquely attributable fault that's one of the reasons we have to do all this data availability sampling stuff uh because you can't you can't prove the you can't prove that data is available to any other node unless the other node just unless unless your proof is just here's the whole data like sure that could be a proof but it's it's not not exact good way of doing it uh so so since you can't since you can't do that you can't do anything onchain unless of course you

just kind of reduce in the worst case the onchain system is literally just downloading all the data from offchain which is not that doesn't work that just doesn't fly uh so how do you how do you slash on chain you need some sort of other committee some sort of other way that is not going to be as secure as the bond to enforce the slashing for instance you could have some secondary token uh that could control a dow and the Dow could determine through some token awaited vot uh to enact the slashing right but then again the security is from is derived from two components one is how much is bonded and two how do you enforce slashing on that Bond and in this case while security isn't you know the large

Bond your security is really potentially this much smaller uh smaller value smaller amount of tokens uh which could have you know fraction of the bond in terms of their dollar value uh that are controlling the Dow and they determine how how the slashing gets how they determine how to slash the bom uh that's just one one way you could do it but you know you could just ask your protocol of choice how else do just slashing you know another way they instead of having the Dow do it they could just have their company do it or like a multier or whatever but then you're just back to reputation like reputation is ultimately what securing this protoc not the bond okay blob defer from this and one

of the nice things of the fact that since there's no tokens you don't have to wor since there not a token Bridge you don't have to worry about you know the problems of you know some token some there's a bug in the in the bridge and after that's you know tokens go rampant and stuff it's purely just passing messages around and not just around but in One Direction uh so how is slashing done in blot stream uh the way works if there's either bug in bloodstream or if whoever is doing the off train work which in this case is the celesti validator Set uh misbehaves uh they need to get they well if there's a bug I guess there there's there could be a penalty there could not who knows but you know this if there's you know misbehavior then they're going to get slashed so who enforces the

slashing uh this actually requir the Celestia validators to misbehave on Celestia itself not just on the brid because not the bridge sorry the relay uh it doesn't just requ them to misbehave on bloodstream it requires them to misbehave onesa itself uh which means they've now corrupted Celestia itself and what happens if you do that well you're going to get slashed so this require so this is going to result in 2third of the Celestia validator getting SL getting slashed and this will be enforced through Celestia social consensus and generally speaking for any moderately mature blockchain you should assume that the cost to corrupt social consensus across a large community is

effectively infinite that basically can't happen right like if you ask the ethereum community hey ethereum can we you know mint a million ether and give it to Robbie right the ethereum community is going to say no and there's not really any amount of money that you could pay them uh to to do that right uh because that would kind of erode like the I mean the amount of money You' have to pay them like it wouldn't be like a million either you have to like you had to like pay out the pay out the woo would be way more than like the amount of staked ether way more than that because it would basically destroy the entire chain forever if they did this similarly you know the Bitcoin Community or the you know Sesto Community basically on any blockchain that has

established a decentralized community the cost to corrupt that community in such an obvious way if you have this obvious very obvious and provable and attributable Corruption to the system uh and then you just let it happen like the cost to do that should be model modeled is infinite and this is very different than for instance one company who has who will have its reputation tarnished right as they're going bankrupt so they don't even care uh or you know or some da uh where the amount staked on that Dow is a tiny fraction of the bond so the guarantees are on slashing in SLA with bloodstream are not even close to comparable to the guarantees or lack the

rough that you get out of any other alternative or rchain data availability solution for ethereum and evm based chains that was kind of longwinded but did that give an overview of lam and it's how was different and and I brought up the uh the graphic because I think that it will help people understand um and also us um but I think I mean understanding the differences between offchain um dac's or effectively just like offchain groups of people saying like yes this is good and yes or no this is not that's kind of like that was like step one I think with like validium and like ear ear early stage rollups and things or early stage

scaling designs and now kind of moving into into slashing and and having the validator set be the ones which verify the truth um and then them being penalized so when that the validators uh are will be verifying in this attestation if you look at at the uh graphic that's kind of like the attestation phase right where the the that the validators are signing that uh that that that the data is truthful and available and then passing it on um that's where slashing could occur is that right slashing doesn't happen in the da attestation in the sense that the da attestation is what is related from Celestia to ethereum the slashing will take effect in the actual data

availability layer itself so it's the little bubble on the right that's a Celestia uh so it would essentially be the social consensus of the celestial Community will determine that there's very easily provably fault in the cestia chain because the validators have to sign over something that is malicious know some some invalid CES block essentially in order to produce an invalid da da attestation uh so this is you know Tri it's trivially provable uh because of that you Celestia no no because we know those paral node and you can verify that the block is the the block is invalid that has been that has been relayed or

unavailable uh so why can we do this when we couldn't do this uh we couldn't have this this nice guarantee uh this nice guarantee a couple years ago uh so when blob stream was first introduced as the quantum gravity Bridge there's the word again we should have called the quantum gravity relay uh a couple years ago or I think it's been maybe a year ago maybe two years ago time flies with Co this time temporal Distortion field uh back then uh we used a very we used kind of a much more naive approach uh that did not use EK uh that essentially uh had there was a situation where validators had two keys uh since

Celestia's based on the cosmos stack uh celestian validators use ed25519 signatures uh but in our theorem to verify those is very expensive uh so the old quantum gravity Bridge used use the situation a situation where every validator had two keys one that they would sign on Celestia another one that they would sign with on ethereum uh and the ethereum keys would use ecdsa with SE 256 K1 similar to you know Bitcoin ethereum and and those kind of chains uh that that use kind of older signature scheme which is still fine it just it is different uh and it's much more expensive to verify an ed25519 signature than you know an ecdsa signature on ethereum specifically uh and you know

the initial Target for quantum gravity bridge and bloodstream would be evm uh based execution environments such as ethereum uh so this led to a bunch of complications because now there's two keys uh you had it was a whole bunch of complications I'm not even going to go into it uh but you the kind of important takeaway is that you didn't have uh in such an easy way this nice guarantee that uh to corrupt the da station you had to corrupt Celestia itself uh Now with uh with now with the new the new scheme which internally we call blob stream x uh developed by succinct uh you do have that guarantee uh so how does blob X work uh it

essentially runs uh it runs a system it runs a protocol where it uh it generates a ZK proof that uh a Celestia block has consensus over it for all celesa block all celesia block headers uh and essentially the ZK proof you know verifies all the very expensive to verify evm ed25519 signatures it verifies them inside a ZK proof and then you only have to verify that one ZK proof for a whole batch of block headers uh on on ethereum at a time so this substantially reduces the cost but also it means validators don't need to have these two different key pairs only need one then the one is exactly the same key

pair that they use to sign s blocks so in order to corrupt the system uh in this with with blob stream X they would actually need to sign over an invalid or unavailable celester block which is much easier to prove than you know if there's an extra key pair involved uh and this protocol uh the nice thing about blob X by syn is that yes while it's currently uh ailable on evm based execution environments because there's an evm verifier it's very easy due to the nature of you know ZK verifiers a lot of them are written on Rust and stuff it's very easy to Port it to for example Salena if you wanted to run it in the Salena VM then you can run uh you know you could have Celestia uh

you can have l3s on salana with Celestia underneath for data availability if you wanted if you wanted to have this uh so kind of to uh to expand on this a couple more points I think a couple weeks ago uh cinct announced uh Cel Labs also that uh blob stream was deployed on base uh so the base roll up on ethereum uh so this means you can build l3s uh like a Base main net so that means you can build l3s on top of Base that leverage Celestia underneath for data availability uh just earlier today today being the date of recording so Tuesday April 2nd uh we ounc that uh blob stream is going to be coming to

arbitrum uh so this will allow you to deploy arbitrum l3s uh with and this is kind of unique thing among all these da Integrations uh the the sorry the these Like You Know offchain da Integrations uh this is working fraud proofs so it's it'll allow you to deploy uh it allow you to deploy arbit from orbit l3s uh on top of nitro if I remember correctly with actual working uh fraud proofs so this it's currently undergoing audit uh but I believe it's going to be in test that either it is now or will be very shortly uh and will be in M

obviously in the in the near future uh so already now and more so in in in the near future you'll be able to uh you'll be able to leverage bloodstream in production on mainnet uh to deploy l3s on top of layer two rollups on ethereum for massively uh increased data availability through put and also massively reduce costs yeah and and uh today's announcement was was exactly that with the orbit chains arbitrum arbitrum Nitro has been pushing the fraud proofs which I think is something that the ethereum community gets a kind of a lot of of a of a bit of a WP for um have have just

couple minutes left so we have some like rapid fire questions in a way feel free to to uh um elaborate as much as you like um I'll go first and I'm sure Rob's got some as well my first one is like does the differentiation of data availability costs really matter that much after a certain level of of scale in terms of orders magnitudes cheaper than ethereum post Denon like is are are we going to see differentiation in teams building out these this uh uh or choosing that their stack based on that these slight differences in cost or are there other metrics or characteristics that we should be looking out for as far

as differentiation amongst these da lers yeah so yes and no uh as as always the case for a lot of these protocol design and protocol choice questions uh so no in the sense that uh all else being for certain use cases you know the difference between a fraction of a penny and a smaller fraction of a penny is well was nothing basically right uh but imagine you have a huge volume uh then even small differences uh can make a substantial difference if the volume is huge uh and this is kind of where the Crux of the maybe no comes in which is that uh

the all blockchains have some sort of constrained resource because constraint on how many resources they can provide to the to the network or to the world I guess it's probably good I I don't know how you want to describe they have a constraint on how many resources are available uh you know manifest is some sort of limit on the through but this could be in terms of you know megabytes per second for modular blockchains uh which one of my pet pees is when someone quotes transactions per second for for like a data layer which makes no sense it's all about you know B per second or you know transaction per second or gas per second what what whatever uh there's going to be but there's going to be some sort of upper limit on these things they don't uh they don't provide you know it's it's not

unlimited for any blockchain uh and that's kind of where the the tricky thing is which is that once this limit is hit a fee Market will emerge unavoidably well it's either Fe Market will emerge or blockchain is going to start just crapping out and just shutting down so we're going to we're going to assume blockchain is not going to do that so a fee Market will emerge uh where users are going to start bidding against each other for inclusion in this very limited block space uh so if you look at two protocols and you say okay both of them are super cheap that basically means nothing because the only reason they're cheap is because neither of them have hit this situation where fee Market will emerge uh

and so like you know one blockchain says you know that million times cheaper than the other one and you know one of them is a fraction of a penny and the other one's you know 100 millionth of a fraction of a penny you know this comparison is disingenuous because these numbers mean nothing because this clear it's clearly not in a fee Market situation so what's the important parameter is actually not the cost uh it's actually the through put uh because the through put is how much how many users will be able to use the blockchain or use this protocol what whatever we'll be able to use this thing before fee Market emerges and potentially the costs become prohibitive so the throughput is the actual important number uh and in that context Celestia has substantially higher

through put than for example post Denon ethereum uh obviously speakes per second not transactions per second or transactions well that that's correct but also in another way so not transactions per second but yes in terms of megabytes per second Celestia has greatly increased throughput which means that you know if they're both used they're both used negligibly then okay it doesn't make much difference but you know as soon as you start using more as soon as you get to a point where rups explode in usage which we we actually saw happen not not not too long ago with things like blob descriptions uh then you're in a situation where fee Market emerges and you know it if you know if one protocol

can do you know one unit of work the other per per second the other one can do 10 you know if you're in a situation where you can do two units of work where sorry where the user demands two units of work you know the first blockchain that can only do one is going to have a fee Market while the second one won't uh so it all depends on where you are on this demand curve but the one with with more throughput is basically universally in all cases regardless of where the demand is going to have the or fees and the throughput of Celestia is much greater than for instance post eum that's the the one thing but the second thing and this is also important is post Enon ethereum allows three blobs per block as a maximum before says no the maximum is actually six that's actually

wrong uh because the maximum is three and double maximum is six uh people always use the framing if oh blocks can be half full and then they can be full but really what it is blocks can be full and and they can be twice full uh which if someone if if someone complains about what I just said on Twitter I'm going to like I'm gonna I'm gonna rip you a new one uh so please do if you think I'm wrong because I'm I am excited to to rip you new one on Twitter uh but basically the maximum the maximum blobs you can have for Block cone is three uh and think about how many RS are in ethereum today and think about how many RS they're going to be right uh we envision a world with you know 1 billion users one gigabyte blocks and one million

rollups that's not going to fit in three blobs per block uh so Celestia not only has it does it have higher throughput but it also has just based on you know Pro some protocol details it has a much more flexible usage of it block space that allows you to have a much larger number of blobs I'm not talking like 10 I'm talking like you can have like 500 blobs if you want they'll obviously be smaller uh but you can have a much larger number of blobs is much more flexible in that regard it doesn't have this restriction uh so in both of those ways uh cestia will provide much more throughput uh than currently exists even in postm ethereum and much more you know for the level of security that it provides than any other offchain data

availability solution for etherum I that's a really interesting mental model about the fee markets um the other one that that I have is uh there's the we had a really interesting uh presentation about monolithic and modular uh structures converging in in a similar path where monolithic structures scale modularly and modular structures are going to enshrine more and understanding that Celestia kind of pioner this modu ecosystem there's now this move towards enshrining a ZK verifier um for an further efficiency and one could call this moving from a more modular to a a more integrated or monolithic

structure why exactly is the ZK verifier um something that Celestia is looking to move towards um I understand kind of like the like the uh Bridge situation in terms of L2 to L2 and L2 to L1 and perhaps you could elaborate a bit Yeah so it's basically entirely because of the brid situation so it's uh so I want to say that Celestia isn't State L I was using that word more because it's shorter to say but the correct response of the or the correct description of the state minimized in the sense that the state machine is extremely minimal uh so sure it's not State less but State minimized uh but you could also argue syn is not

completely State less it's state minimized then it's not you know oh it's not some pure ideological like pure version of the leder white paper which is true uh so it's already it's already One Step removed from that it's already one step towards you know you can call it some settle if you want to call it that I mean that what the discretion means nothing at this point uh but the point of Stark accounts It We call we call them snark accounts uh name subject to change uh we're trying to find we're brainstorming some cute names uh there there are some there are some bad suggestions there are some good suggestions uh okay there's some bad suggest the medium suggestions I don't think we found a good name but we'll call them star accounts or the ZK

accounts uh the purpose of them is to allow for secure two-way bridging between Celestia and rollups in this case it's bridging so it's bridging of tokens uh such as Tia or if you happen to have any other token representations uh most of here uh between like two and from Celestia and roll ups on last year but the reason that uh but like both the proposal and the reason for the proposal is to enable this bridge without harming substantially the trust minimization nature the trust minimize nature of the celestial State machine sure you could say okay well you know let's go full evm let's build the whole t t complete State machine now you can Bridge tokens to and

from you know smart contract Bridges but the issue with that is then it's not stay minimized anymore so SAR accounts are this nice compromise where we get TR trust minimized bridging uh with certain constraints obviously you can't have fully expressive message passing it would be you know something simpler like a token Bridge uh between slan Roll-Ups and this kind of greatly reduces the friction of running rollups and Celestia and paying fees and tia uh etc etc it reduces a bunch of frictions uh both between rups and between rups and Celestia and it does so while not harming the state minimization property as much as possible right because snarks are kind of like a thing where you can express these complex computations with very little complexity on the Bas layer

because the verifier is much simpler than what's happening inside the actual proofs awesome all right one one last one I know that we're super low on time um you guys chose a model for data availability sampling and for data availability that wasn't um around kzg commitments I understand that you guys use eraser coding um however this kind of kzg polinomial math verified uh verified liability for for my terms is it what you guys chose what's the trade-offs why why is that design Choice why was that design Choice made uh primarily due to time in the

sense oh well there's two reasons one of them is primarily due to time and the sense that Celestia is live on mayet all the protocols that have attempted to go the kcg route are not uh so that you know tells you everything you need to know with respect to time to Market uh and like it's not just that they're not you know but they're a couple months behind is you know Celestia has been on mayet for almost 6 months and nothing else is even close uh so that's the first thing uh time to Market was substantially faster uh and it could be done without any you know compromises with respect to the cryptographic compromises or whatever to the fra proofs or whatever you know the bad Co fra proofs and stuff are all there uh so like data availability something has

this security that you get if you read the paper with respect to the existence of fraud proofs it's not like some protocols or fraud proofs don't exist so you know it's it's all there uh and know you know Shi clearly substantially faster than kgd commitments so at the time that this was being developed uh kgd commitments were very immature both in terms of general knowledge about how they work cryptographically and also the proven costs etc etc uh and also how to verify them on ethereum right like a lot uh if you've talked to any of the zekk RO teams they'll tell that integrating K commitments into those Z rollups verifying that verification is non-trivial it's a lot of work right while you know verifying something or integrating something in

bloodstream is substantially last work right uh like it can it can be done with substantially work potentially faster so that's kind of the first point uh the second point is that you don't like you don't really get substantial benefits out of kgd commitments some people say oh you remove the synchron assumption you remove the need to wait for some honest party you don't need an honest party but that's not really true because even with kcg commitments that's only for sampling yes kgd commitments remove the need for bad encoding for proof during the sampling process but they don't remove the need for reconstruction and like I said there's two parts to like the whole like data availability

sampling protocol one of them is sampling the other one is reconstruction and even with kg commitments you still need uh you still need nodes to be able to do that reconstruction uh which which also requires a synchron assumption so you you can't remove you can't remove entirely the need for multiple noes communicating with each other over a synchronous Network and therefore you can't remove the assumption that there's other honest nodes in the network entirely because of that second part and that second part has nothing to do with ZK it just is there can the nodes and talk to each other to reconstruct the block uh so because of this thing is like don't get like yes you get some benefits of kg commitment sure you don't get you know

you don't get a thing where it's just strictly better in every way I guess it is strictly better in every way except for the you know obviously the cost to develop but you don't have a situation where it's like you know so much better that it's you know you have you you have to go straight to that and that was is not that that is not the case uh and of course once Cas Commandments are mature I fully expect the Celestia Community to endorse and approve of an upgrade to the Celestia protocol that will allow migration to KD committs and we thought a lot about how we can update the protocol without while also retaining backwards compatibility and other words while not harming you know existing rollups and all protocols that are building on Celestia uh so you know that's you just you just upgraded to use

KJ commitments for something so there's always an upgrade path later uh and it just wasn't the blocker toand and launch got it all right man well I I I I very much appreciate your time and I I know Rob does as well appreciate you coming on I have plenty more questions about Sovereign rollups as well but um I think I think we can we can save those for next time Rob yeah sounds great playing awks yeah thanks for having me on that was great awesome cheers