# From 4844 to Danksharding, a DA layer roadmap | Francesco Damato (Januray 2024) - Channel: [Berlin Ethereum Meetup](https://streameth.org/berlin-ethereum-meetup) - Date: 2024-10-07 - Duration: 46:30 - Watch: https://streameth.org/watch/yt-eCIPCh2b5aU - YouTube: https://www.youtube.com/watch?v=eCIPCh2b5aU ## Description Join us on Meetup to keep track of our events in Berlin: https://www.meetup.com/berlin-ethereu... See you at the next one! --- Apply to speak at our future meetups: https://forms.gle/5y9Y5ywZC7pSEqpV9 --- Twitter: @BerlinMeetup ## Transcript hi everyone um I am Franchesco I'm also I work at thef well I'm a researcher and uh I decided to talk about something which I think is quite um kind of relevant to today because I'm not exact to this period um because was probably a lot of you know in about a month uh the D hard workk is going to activate and with that is going to come e444 which is a pretty dramatic change in the data availability uh layer of ethereum and also stage for future uh even maybe more substantial changes uh so the idea of the talk is essentially to be a bit of a where we are coming from where we are and we're going in terms of data avilability um later and uh yeah so the starting point was actually one that had nothing to do with data availability in some sense is just this uh single blockchain some might say monolithic blockchain uh where essentially everyone downloads their whole data and then you blocks just contain trans transaction yeah blocks just contain transactions transactions have these little like Blobs of data which we call call data that essentially tell you all you need to know to execute them you just execute everything and that's that's the whole story there's like data really doesn't play kind of a special role here um but that has already changed quite a lot in the last few years um so we've already entered this kind of new paradigm of the rollup Cent Ro map where instead of just having this one ethereum blockchain we have whole bunch of other blockchains that are essentially built on it which we call rollups for example like arbitrum optimism and a bunch of others at this point it's like too too long of a list to the name and uh the way they work essentially is that they utilize the E they utilize eum transactions and especially the call data beher transactions so the again these little blobs inside the transactions to post their own transactions um and ethereum basically doesn't execute those transactions but uh just essentially checks the validity through some other means like either optimistically or actually through ver final proof and this is kind of where the scalability comes from at this point um and yeah so zooming in a bit more you have these blocks of data that are contained in transactions on ethereum in this case they're ones that have to do with optimism they themselves contain other transactions that are the ones that optimism wants to use for its own chain um and essentially how you figure out what's going on with the optimis and change is you read the transaction transactions and execute them and get the chain um and yeah basically the whole picture is kind of like this like the this rollup chain essentially just reads from cold data on ethereum and that this is the way that it figures out what its own state [Music] is um but in Practice Things actually look a bit different than the picture before um yes we do have these you know arbitrum optimism zzing whatever transactions with their C data on ethereum but actually most of ethereum blocks are full of regular transaction L1 transactions and the reason for this is basically that call data is priced in the same way as all other resources that we have in theem at the moment which is gas um like anything you want to do with a smart contract basically anything you want to do on etherum you're going to need to pay gas for and basically L2 transactions are competing with one transaction for gas essentially and it's it's hard for them to do so because L L1 transactions are quite valuable so essentially this the outcome that you get is that um you would need a much bigger demand for rollups in order to actually fill up L1 blocks with rollup transactions um because of this competitive effect and the other thing is that we couldn't make all data cheaper U so easily because basically right now everyone keeps this call data forever essentially like you you keep the whole is of the chain and it's something that people have proposed to change but it's a bit it's it's not so there's some push back on that and things would have to kind of be put in place in order to be sure that the history actually is retained in order to do that um and so essentially we have to take this into account that when cold data is put on chain it's going to stick around forever and so you you don't want to make it too cheap essentially and yeah you can actually see that in practice this effect because uh in theory ethereum has about 1.8 megabytes per uh block of cold data maximum but in practice what's used is like about 140 actually much less than that kilobytes so it's like about 1 GB per day you can see in this picture it's like a maximum is actually never really reached uh usually it's more like 500 megabytes or something like that and uh which corresponds to about so 1 GB would correspond to about 140 kiloby so still like more than order magn to less uh than the theoretical maximum C data that theorem uh would potentially could potentially have in Block um and yeah basically this kind of sets a stage for um [Music] this ass stage for the uh the next upgrade IP 444 um so the way I 444 improves the situation is basically twofold um is one is that it introduces this new data resource so we kind of start doing a way with call data at least as a resource for rollups um and basically this data resource which we call blobs as its own fee Market which is really an important thing we kind of want to get rid of this competitive effect um and then we do Easter expiry just for uh blobs so not trying to get you know this complete um history expir where nodes really just forget about the the whole chain after some time but just for blocks so a bit more restricted and kind of easier more palatable essentially um yeah so basically the uh what blogs are is essentially just some bunch of arbitrary data 128 Kobes each there's details to what they actually are but it's not really super relevant here and the kind of important thing for at least the purposes of this talk is that yeah they have their own uh fee market and it has this uh e559 style uh Dynamic pricing um that really just works for blobs themselves um and this basically lets us set exactly Target of three blobs and a limit of six uh per spot or I mean whichever Target and limit we would like and this essentially means it so that we can really regulate how much capacity uh we have and because it doesn't have anything to do with the um market for gas it's really we can really make sure that it's actually used so if we set this capacity aside this is the capacity that's going to be used essentially um and yeah then also we have Easter expiry so now noes essentially only have to keep this data around for about 18 days which really limits the amount of uh storage space that they need for this um yeah per day basically with thep 44 4 we'll have it Target of about 2.7 GB per per um per day of blob data which is compared to what we saw before so around like 500 megabytes let's say per day um at the moment it's like about a 5x increase in capacity essentially um which I mean obos it's very good and we'll see actually in a month how it this actually looks in terms of fees and and everything um but there's this kind of leaves a problem that e 444 doesn't actually really do anything in terms of um making basically going Beyond just depending on the bandwidth of individual full notes so we're still completely bottlenecked on how much can an individual food not handle in terms of bandwidth because everyone is still downloading the whole data so we like created this new data these blogs but everyone is downloading every blob and we only get the scalability the additional scalability from these fee Market changes essentially in the history expiry and now we want to do more than that we want to make sure that we are not actually borrow anymore by the band within individual notes at least less than than now and yeah you might have heard the term sharding and if that tells you anything you can see that there's nothing really sharded up to this point um and yeah here comes essentially uh the the next stage of what you know we essentially would like to uh to do at some point after uh the the next work um which is data availability sampling which has been the goal at this point for for years and at this point uh the idea would be to do what I'll explain what it means but a onedimensional version of data availability sampling so the the idea of dat availability sampling is somewhat simple it's it's that essentially we take a blob which will be the the blue part here the the left part and then we essentially double its size in a in a special way not just in in in any way but we do it in a way where and we call this an extension where we basically introduce redundancy this is also called erasor coding um so that as long as any 50% of it is a u as long as you have any 50% of it you can reconstruct the full blob um this basically means that if you want to know that a blob exists is is available um you only need to make sure that 50% of it is available which changes the problem quite a bit and so the the way this uh actually changes the problem is that uh now you can download only like a really small amount of The Blob you don't even need to download 50% of it you only need to basically randomly pick a few uh points in the blob uh to ask like you ask someone give me this this point this point that point randomly Chosen and if all the points are are actually available with high probability the whole thing is available and so the the nice thing here is that you only need to download a constant amount of data essentially to make this check yeah and uh basically the so that's that's for one blob uh but we have potentially right now three from 3 to six but it could be more um the way you would do it is that with more blobs is that essentially you do the same check but for all blobs at once so you essentially pick some indices and then just download like a column like you you stack blobs on top of each other and then just um a sample becomes instead of a point on a Blog a whole column this is essentially equivalent to just doing the check for each blob independently but where you fix the randomness uh once essentially um so this essenti it's quite simple it's still like a 1D uh one dimensional construction but it's uh allows you to check for all of the ones and uh yeah the the important question at this point is like okay sure if like magically you can get the samples you can make these checks but how do you actually get the samples and that's been I guess one of the main questions of research around this area for years at this point um so yeah the idea at the moment is to do what we we're calling peer do and the peer part is essentially tells you how it is that people actually get samples um so there's been like a variety of ideas around this area um so essentially uh the the IDE of here does is that you would just um maintain all Network infrastructure that we have today and just reuse it to do sampling and this is kind of a simplification over a lot of other ideas that have been uh put forth like for example dhds have been investigated for years and they also have their own problem but it's also a question of just complexity as opposed to just trying to reuse networking component you already have so the idea what here does is literally just to do what you might think is the simplest thing possible which is just ask your peers you already have some set of networking peers um that you keep for other things that you need to do uh in the in the like just in the consensus layer Network just ask your peers uh for samples with some way of figuring out which gears are supposed to have which samples um so this essentially mostly simplification um but this doesn't kind of answer every question that we have because still how do these fears how these nodes that you have around you that you're supposed to ask for samples how do they actually get the samples is also very important part of the question and this you might call the distribution problem um so somehow the data has to be like disseminated in alw in the first place um and we want to and want to avoid this being like another kind of M link and uh kind of at high level I do might think about distribution is we want to make sure that everyone only downloads a small part of the data as opposed to the whole data again that's kind of the principle of the whole thing um in this case it's not going to be a constant amount of data like sampling unfortunately that's sort of unavoidable um but for example right now we think that a good way to do this might be to have uh nodes download only 13 second of the data at a high level it looks like you have some big piece of data it's going to reach a bunch of blocks ultimately you split it up in a certain way you assign um basically you partition the validator set in the same way so let's say in 32 uh subsets then you assign each subset to a uh to a piece of data and just somehow people are going to get the um the the sign piece of data um but that's just like at a high level then how it's actually going to work in practice is again trying to reuse things that we already have so not trying to uh basically utilize new networking components or constructions but just trying to reuse things that were already kind of comfortable with and already well tested and clients already have essentially um for other reasons in in subnets which uh you can basically think of subet as just there exists the the whole etherum network and then there are these um basically sub networks where people are connected only to some of um the the rest of the network instead of all um and they only use them for specific things so in this case there will be like 32 of these top networks that each one dedicated to specific part of the data um and so yeah essentially going back to the the picture before we have a bunch of blobs they've been extended in this way to um do this uh for the sampling part and then we split the data uh in some way like it's going to be 32 chunks each chunk is going to be essentially a bunch of these columns which are ultimately the samples that we care about um and yeah basically as a note you would have to go in a subnet that corresponds to one of these big chunks so four columns um essentially are assigned to you um and just essentially gossip about that without notes the take part in propagating that and then essentially have custody of that and be available to respond to sampling queries related to that data um so once you have distribution so once you like somehow the data has been Associated to a block it's been disseminated in the network everyone has their their assigned part if you want to do sampling what you would do is you figure out let's say that you want this uh first red column on the left because you've like randomly flipped a coin and figured out okay this is my first sample you figure out which for your peers is in that subnet uh which is something that you can know from their not ID and essentially ask them just to give you the column um and you do it do it for all the samples essentially so all you want to make sure of is that your Pier set is basically diverse enough in terms of covering all the all the subnets and uh yeah another question that kind of so we've covered essentially distribution and sampling which are mainly two the of the main aspects of the the whole thing um another aspect that's quite important is reconstruction so the the whole principle is that if the data is more than 50% available we basically treat it has been actually fully available so if someone really wanted to um get the data which which might be useful for um you know running a fraud prooof um like a challenge essentially for an optimistic rollup or if you want to make sure that um a validity roll up um is live so someone has the data can uh push forward the state um so there might be cases where you actually really want to ensure that the data is not just 50% available but it's like actually reconstructible to get the the full data and um basically the idea is that eventually we would like this to be a prey robust distributed property something that the network could do um as a whole collectively let's say um so we're essentially any node can contribute to this reconstruction part uh for example by constructing a single blob as opposed to the whole kind of bigger Matrix um so essenti data would be you shouldn't need to touch the whole data to to be able to do reconstruction you can only do it you can only contribute to reconstruction just with some small parts of it that doesn't cost you a lot of bandwidth or computation effort and so on um but yeah for now the idea is that um again in this mindset of simplifying and trying to do things gradually so uh not trying to build like the perfect protocol at once but like kind of going one step at a time and introducing more and more components over time which has been in some sense the philosophy since 4844 as well because 4844 has already built a lot of The cryptographic Primitives for example uh and and other things as well um so yeah basically J would be for now to uh not try to do distributor reconstruction but instead to essentially rely on uh one of an assumption which um for example is like something similar to what optimistic rups do um just assuming that there is some node somewhere that is willing to um basically um follow what's happening and and make a challenge in case there is fraud so the idea here would be somewhat similar onean assumption but for livess we're essentially assume that that at least someone is downloading the whole data and they'll be willing to reconstruct it if needed so it could be like any one of you in the room could run and know that downloads the whole data especially at first because it's not going to be that much data and if needed just rece Network or it could be like an infrastructure provider or like essentially there's a lot of parties that might have reasons to have the whole data and we essentially rely on at least one of them to kind of honestly do this uh this job of reconstructing and yeah this is only really important for livess and not for SA in the sense that Offline that anyone could go and put the pieces together and do reconstruction but we kind of need someone to be online and doing this uh in a timely way for the network to be live and yeah um so with first we saw that we get maybe around like a 5x uh increase in in um uh essentially data throughput or which corresponds to like rollup through it at the end of the day um which is quite nice but again we want to do more and uh with P do essentially without changing anything about the bandwidth that basically is set by fre for 4 so getting like a bandwidth is similar to uh what we get with the three to six uh Blobs of fre for 4 we could essentially have uh 32 to 64 so like a target of 32 and um a Max of 64 uh block s slot which is around a 10x increase we probably wouldn't actually immediately just go in and and and just bump up the The Blob comp by the X but this could be like a gradual thing um maybe at first just go to 816 and then slowly like get there um and but then there is a point where this becomes also Limited in itself where like essentially to get more scaling from this you would have to essentially um again increase band with requirements like it's it wouldn't just be possible to keep bumping up the block count without actually uh going over what for for four requires um at which point one thing that we can do is to do what essentially originally was proposed as for um as like kind of the the final g game of the availability sampling which is this 2D uh construction that you might have heard with the with the name on Shing but you can also think of it as 2dp do essentially uh where we basically do all we have before is is the top there um so blobs stacked vertically and extended horizontally but then there is another exension like a vertical extension so we introduce even more redundancy in the data in this case vertical redundancy so that even columns can be potentially reconstructed uh if like only 50% of them are available um and uh yeah basically um what this does is that sampling becomes a lot cheaper because you don't need to download columns anymore um but only you can only download cells um so essentially sampling becomes like negligibly uh cheap in in in the whole computation and only distribution becomes the the you know the determining factor of how much F people actually need um so yeah that that would be kind of another step uh but essentially we the the nice thing is that we're quite far from actually really needing to do this um so we can take one to Fe D like pretty far um and oh okay um yeah and maybe with this I think uh we can without increasing band get to uh even uh 256 Max blocks which was the original kind of Target through for D Charing which will be around 4X from 444 um and yeah that's that's it uh you can find more on the post on the right which is from Danny uh it's like the original post that introduced here D research and this other post that I recently put research which is somewhat similar to this presentation but cing slightly different things thank you and uh I [Applause] think okay yeah okay then just later I guess yeah questions if you have any yeah we can do them now when will this happen um what's the yeah I mean I think it's not really on me to give timelines I don't think anyone likes to do this um I would like this to happen as early as possible because I think it's very important like to me scalability is like a really big priority in the the road map I think in principle because of the fact that a lot of the design uh here really relies on reusing things they already have like cents already comfortable with there's not so much more complexity I think maybe the the um stickier point is that uh still it touches a lot of networking things which tend to be more complicated than others like a bit I guess something that we're less used to but also I think wither for for a lot of that has Chang a lot of like Tes infastructure has been built around and you probably talk to par there in the backb this definitely knows a lot more about me but yeah I think so March uh next year you mean uh yeah I mean I I think possibly like it's not like it's not an thing but yeah it's it's really going to depend but also it depends what exactly means what's the next step because the next step could be a lot of p d is built but we don't actually increase the block count just for example we could switch over to using sampling but without actually increasing the number of blobs too much so that it's kind of a control switch um I if you have was I don't know whether you mentioned it I missed it but is that going to be an additional incentive system and penalty system or is it built into the current uh proof of stake incentive system um system for to keeping like if you don't have the blob right you you're being asked for the sample and then you answer with a timeout because you actually don't have it right like so I think it's it's very hard to incentivize this kind like serving data because it's you know how do you know that someone has done it or not it's it's quite hard what you can do is um on the NW workking level you can introduce um basically just use beer scoring techniques that we already have like a bunch of for for all kinds of things so like if if your peers are consistently not able ble to give you samples that supposedly they should have you can downcore them or things like that you can you can look for different peers and so there's a lot of I think like networking level management things that that you can do um I I don't think there would be like an one chain let's say like incentive reward system for for this um eventually there could be something like proof of custody which is a bit of a whole different thing um but that's more just for basically making sure that there is an incentive to not be a lazy validator and if you're being assigned um a certain piece of data to to download you actually have incentive to to go and download it instead of just signing off on it but here it's not even I mean when I I guess all that I've talked about here it doesn't really have anything too specific to validators it's really like all full nodes we need to be doing this we need to be serving data participating distribution so on so it's not something that depends specifically on validators but basally we do have the option potential to do something a bit different for validators and make I guess make um their r in the in the scheme potentially stronger and a bit uh yeah treated it differently with different guarantees um as far as I understand you don't have access to this data from the solidity smart contract is that correct yeah so so already from 344 you only access the the commment yeah is there any other way how as a the app developer smart contract developer I can use this data how can that I get access to it I mean there is the point evaluation precompile which is like a precompile that allows you to basically yeah get like um uh basically prove that the data at a specific point is what it's you're you're telling that that it is but uh yeah you will not be able to directly just wear the data like basically if you want to use that data you have to like provide it from outside as called data and prove that it is what to set it is but I I think really I mean I I don't know gu I have never thought about how can you use this data outside of robuxs in my mind this is just a thing for robuxs uh and it's not something that should necessar needs to have application for for regular contracts and those should still be using C data but I could be wrong wondering if it's possible to abuse it sort of for storage or something yeah I don't really know it not something that I personally have thought too much about but me if there anyone else in the room has thoughts about this feel free to speak up um just to summarize again so 44 you said LS are coming we reduce the cost for RS and then basically explaining what's coming next could you just say so um initially you start how many bus you say like just six 16 so six is the so I mean originally when the VIP was written it was meant to be a target of in the max of 16 now it's a target of 3 in the max of 6 which correspond to about 37 mbes of Target and like twice as much as a limit but initially the only start three but just already with the ma uh will start with three yeah like oh um I see I see what the confusions so when I say Target a Max the max isn't like a value that we eventually want to get to it's just that in this kind of 1559 style uh pricing few Market mechanisms you have basically a target for the utilization of a resource which is like I don't know in the case of gas it's 15 million gas which is like essentially on average how much you use uh and then there's a Max which is in any given block you might use as much as 30 million gas um but on average you cannot have 30 Mill in this case it's the same I mean to be honest with you I don't really understand but let me ask in a very simple way there a lot of calculations that say Hey after after the C uh so much and so much like cost savings for gas cost what do we think some of them are super optimistic and some of them are not so initially what we will get so just for you because like is it like more Super 10x or maybe just 3x 4X and the lat I think the issue is so what we can say for sure is that compared to today roughly there will be about five more rollup transaction just because the capacity uh like the target capacity is about five times more what it actually use in practice today I me today there isn't a Target the amount that is used just depends on how much it costs and how much people are willing to uh to pay essentially um so the the capacity will increase by 5x what the fees do is a different matter because it could be that the capacity increases by 5x but because of all this increase and you more activity people are actually willing to pay more and so there's 5X transactions but they also cost the same as today or like I mean principally could even cost more but probably not like you know what's like going to be the case is is somewhere in between I think so that yeah we do get this increas in capacity it does reduce fees but probably not like by by maybe by the amount that you might expect because there is some kind of induced demand and yeah it's just it's not really that simple but we do know that the capacity will increase by that much like that's for sure sorry I've beening um okay so regarding the P um what is the orchestrator that decides what nodes are serving what Slots of data and I also okay um so basically it's it's all decided by Den not ID so yeah Den ID will just um through some function tell everyone else that that easier clear which uh of these like two subnets you've been assigned to and so you know that that tells them exactly like if I need a call the subnet I should ask this guy because is supposed to Happ okay so then there yeah there's a level ofm to are supposed to be online constantly serving these serving these like call specifically um and then the other the second part of the question was when whoever came up with P do uh and like the the need prepared us was there any kind of like metrics done on just having every single every single column or was this just an idea that came out of like oh one day we'll probably need this like was this was this like a metrics based decision or was this more like a research based so you mean you mean like as in could we just have we checked if it's actually necessary to do the splitting yeah I would say yes because just um I mean for one thing a lot of tests have been done for for 844 specifically um because it does already have this element like increasing how much band uh nodes use and yeah partially as I initially as I said before like it started from eight blocks being the target it's been reduced to three this because there's actually been done lots of test that been done on Main net for example with Big Blocks or things like that just to see like how does Network handle a lot of data going through and it handles it okay but like not like we clearly we could not have you know 10x as much data going through Network I have a bunch of followups some maybe I should ask them later but um they're more along the lines of like do you guys use Li transport for this specific component um so is it some of no it's yeah it's using it's really using the same like Li P2P things that that we use for the rest of the the the consens layer I should say it's interesting it would be really nice to understand what where the strain actually is because you're box are quite small so they're they're quite small the block times are like what 12 seconds yeah and so Al I should say like the block times are 12 seconds but the propagation needs to be like it's four actually okay well regardless um the okay so the block time 12 seconds you need the 4 seconds that but the the through is actually quite small like you you have quite small blocks and then also very small block so it would be very interesting to understand where the actual strin is coming from yeah I mean I think partially is uh that there is this uh amplification factor that comes from the way blocks are propagated like uh is like through I mean gossip best is like gossiping yeah and and so yeah you essentially have I think it's X amplification effect we get um soip yeah like the vanilla gosip impation uh yeah it's like sub 1.1 okay um so yeah that's part of it I don't know if like I mean yeah I think also like block sizes don't really compare too well to like blocks are much smaller than what blobs would be like even even uh I think I know the amount of cool data on average that is in blocks is less than 100 kiloby kilobytes which is not even one more so it's already still quite a bit yeah it's defitely small but yeah it's it's I me I think a lot of it is is just that we're still kind of quite interested in supporting a wide range of nodes uh both for sticking and for just like running and a lot this it's not just for staing but like fors as well but I'm sure yeah it's very possible that there might be like optimizations I'm not really a working person myself but and yeah just had a brief question um so you mention the subnets um through the subnets you can know which node stores have data so you can actually reconstruct the data but where are these node identif actually stored like how they can access where are these node identifiers so you mentioned through the node ID I actually know which other nodes have specific kind of I'm looking for so how I like how do then if you run through that process of me reconstructing the data I need to whatever request how would it look like um so I guess I'm not entirely sure I understand the Reconstruction part of your question so how I mean if I were to have a request and I to a certain piece of data where it requires the dat now that different kind of notes are storing because you're saying there data Avail sampling how do I then get through the note ID the data I need for whatever request I have so basically like you know the no idea of repeaters and from so I mean if you're getting a request from someone it's because by your your no ID you're actually already supposed to have the data is supposed to needing to go in ask someone else like if your your no ID dictates that you're in subnet one you and then someone asked you for a column that belongs to subnet one you should already have the data and the way you should done it is that being in a subnet essentially means connecting to other peers that are also interested in the same topic so are also in the subnet and you would have participated in propagation of um of this um of these columns that belong to the submit so by the time someone asks you for it you should already have it as far as like how do you find nodes that are in a certain subnet if there is we already have like Discovery um processes that that nodes use to find other nodes and that that would use the same similar infrastructure um and but like at a certain point you shouldn't be doing this all the time it's not that you let's say toss a coin and say okay this time I want column 10 let me find someone else column 10 you should already have set of geers that you're like Ste be connected to which you already know cover the the subnet space well so that you can just point at someone that know ask all tenant just ask them for it so you shouldn't like constantly be running Discovery to to find new notes as like live kind okay um is the sampling process part of the consensus or like where does that come in um so yeah I guess something I should have said is U all of the basically all of P does is actually just the networking upgrade the only part that actually touches anything else really is is the block count increase like if you want to increase the block okay you need to like do some other stuff but other than that it's really just a networking um change and in principle yeah you don't really like I mean the beon sh knows nothing about this there's no there's nothing that has to do with like state transitioner or it's all kind of outside I mean obviously like you need noes to um if you want notes to speak to each other they all need to like be on p p but like um yeah I don't so when starting to prune here um would you say makes sense to prune the rest as well so just proving the entire blockchain after all like with G for example you can verify from Genesis anyway right I mean personally I'm Pro proving because I think that given that we have subjectivity as like already kind of part of the design it it makes it doesn't really make sense to me to to sing from Genesis um but but I do understand that like there's a some level of like anxiety about you know really changing in this fundamental way like we're just like having a strong guarantee of like the whole blockchain is like replicated in all of these nodes and it feels nice and you can go and look at the whole history um I wouldn't particularly be worried that the history would disappear but yeah I think there's like a lot of differing opinions I think this is a lot more contentious than than pring blobs I think also just because basically pring blobs changes nothing for you if all you care about is eum L1 like even if Globs disappear if you don't care about rollups you basically just don't really need to care um and even for rollups it's like fine in the sense that at that point they whatever like any kind of fra proof game whatever already happened by the time 18 days have passed so I think it's it's a bit of a different decision how does the approving of this work for these BLS I'm to know that my was actually includ um um so for T here ol I think there's like some kind of a proof of equivalence um thing that you need to do like within the like you have a validity proof and inside the proof you're going to need to prove something about the the commitment um for optimistic pups I don't think you really yeah I don't think there's like too much that would you need to download entire like every single blog or would you just be able to download own BL and then have like I see um so I the yeah I mean ultimately as a rollup node you should only need to download U the data that's relevant to yous are relevant to you but I think you would need to do that somehow like through Network as opposed to like directly being able to um from L1 like point to this I mean you can through like request response it's really request like specific blobs that you want but like for example if you're trying to just live like follow what's happening you wouldn't know that a blob that's relevant to like arbitr or whatever has been included at a specific index so it would there isn't like a clear way for you to just like follow your blogs um that or like you have to rely on like someh the you get to the network and then you just get it from there um like B some like bridge that only but Bridge not in the sense of like smart that justs at that yeah some know that it's like downloading everything just Dum only those blogs in the L2 Network or something but yeah I mean I think um these are also good questions letting this parts of the design are can be changed uh if like I don't know different decisions make more sense so um I think we're like open to listening to like try to work with other tools to figure out one Mak S I mean just to connect what you just said like not technical question has this been validated with the PE at arbitrum and optimism so they can actually use it so I mean so from their perspective I think a lot of it should really um a lot of it should not really change after 4844 like they'll need to upgrade uh to 4844 like to support blobs but then p d won't really change that much for them like the only thing that what it should look like from this perspective is all of a sudden there's more blobs basically um I mean that's not quite entirely true because yeah there is this aspect of okay how do you actually get the blobs from the network um and yeah I think that's I don't know I would say it's like TBD I don't know exactly what's like the the the intended method for for them to do that um but they reviewed thep or they feedback on that or um I mean I not really sure person party has oh do you mean like 444 or okay yeah yeah no 444 yeah absolutely yes I mean they're like in the process of like upgrading their Tes and stuff um but uh yeah P I don't know it's there is it's been shared as like a research idea I don't know if like they look deeply into it to be honest but um yeah hopefully they will how much do you think nowadays compared to past years um I mean I guess depends what you mean by influence I think I I don't find it strange to focus on this it's not like all rups are like begging us to focus on this because this is literally what the road map P has been I want to say for the last however many years I guess three four or something um depends how you think about it maybe you could say that it's always been the road map like it's this is like some form of Charing at the end of the day so yeah I mean I think it's it's also healthy in some sense that they do influence the road map but I I wouldn't say that this is a case of like oh we're doing this because the are I don't know like it's just we do want etherum to scale and this is how we've chosen to scale it so let's scale it and that's my view but and put whatever I want and uh so it is accessible to write this stor what about like restrict access to to write no it's it's permissionless like everything to to make a transaction of of the format which includes a kcg commitment to a blob so like basically there's this transaction format which contains specific type of transaction yeah but it's not like it's unrestricted to anyone so like anyone transaction just pay for the blob and yeah it's just uh but but I mean be just cheaper um yeah yeah I I again I don't know what exactly you would do with it outside of the rollup but I'm sure you can find things yeah no yeah sure yeah no you can I mean you can yeah you can it's AR it's it's arbitary data it doesn't have any like interpretation so a priority like it's only protoc on top that would interpret in some way and it could be a roll up it could be something else so I mean you can even do that on test right now if you you try on on test Nets on on curly already for and is for 3 hours or something and um [Applause] question