MODULAR BASE LAYER with Anurag Arjun of Avail

Transcript

welcome to the interop my guest today is anag Arjun he's the co-founder at Avail their data availability layer for modern blockchains in today's conversation we will discuss the Avil Genesis story and how it spun out of polygon we'll dive into the different components of the modular stack and the emerging ecosystem of teams in the modular space we'll look at ail's architecture what makes it unique we'll also get into Data availability sampling see how that works and talking about the economics and tokenomics of the platform well I'm also dying to find out why he thinks app chains are making their way into ethereum and threatening evm dominance before we get started make sure to hit the like button hit notification Bell And subscribe to get notified of new

episodes every week and remember that none of what we discuss on the interrupt is investment advice and if you enjoyed this content please consider sticking with us we're live on OS Quicksilver osmosis Juno and noas just look for interupt in the active set my guest unag is coming up next right here on the interupt [Music] hey everyone Hi Seb U thanks for having me on the on the podcast and you know like really looking forward to this discussion quick introduction um um my name is anurag I co-founder of aale it's

a is a new kind of layer one blockchain that focuses on providing uh data availability to rollups on top uh primarily over the last 2 three years what we've seen is this dominance of the rollup execution Paradigm with execution moving from the base layer to uh the layer on top and this we've seen primarily on ethereum uh but um what what these rollups really require is you know like tons of data availability and so that's that's kind of the purpose with which we started uh aail uh back in late 2020 uh just to give some context um on uh the initiation story I co-founded polygon a Matic it was uh back then in 2017 that's

actually when I entered the uh crypto space um before that I was a more fintech Rec Tech kind of a guy uh and uh uh so this after we kind of launched polygon POS back in 2020 mid 2020 uh that was when we were kind of looking for what to do next on the polygon road map and so we had execution on our mind and datability as well in the sense that uh Matic really started out as a plasma scaling solution which was primarily an offchain execution and offin data availability solution of course plasma never got implemented uh but you know like we pivoted to POS first and then we

we knew that P was not going to be a long-lasting solution so so that's why we pivoted to GK rollups on the execution side and data availability on the aail side and so so that that's that's where really you know like um um the uh AA really came from this need this understanding that rollups are going to be the future and that these rollups will increasingly need and the scalable data abity layer because rollups are very data hungry um and you really need a base layer that is specifically designed for that uh in in contrast like exe ethereum really began Life as a more execution focused chain and is now pting to data

availability um and so so that's why really um ail really came through um we've been working on aail since the last two and a half three years now and in March uh earlier this year we kind of spun it out of uh polygon and it's now its own separate independent entity with its own team we' have actually now grown pretty fast um like when we spun out we were 20 people strong 18 20 people strong and now we are actually up to 40 U most of mostly are Engineers uh but we are um you know like growing the team U as sustainably as possible yeah that's really it's really

impressive uh that uh that uh you were able to sort of spin it out and and grow the team to to such a a large number um like when when you were uh working on um on aail when you were still at polygon uh before it was spun out like were were there areas where you felt like the polygon Vision was falling short and you felt like a veil could solve uh some of the issues there or was it just sort of an ideological split about how to scale blockchains cuz like you know polygon has now gone this whole like ZK uh route for scaling uh and building roll ups and and I mean that's a very uh

different approach to the data availability approach so yeah what's the what's the chasm there that like caused the veil to want to like go in a different direction yeah so I um you know like uh unfortunately or fortunately like we didn't have much of a you know like a rift or a km in in that in the literal sense mostly it was primarily because uh like if you kind of look back um uh at polygon's product Suite right like it's pretty pretty wide so we've we you know at that point in time we have had like a polygon POS but we also had like a polygon Hermes zkm a polygon zero solution polygon miden solution of

course polygon aale then you also have polygon ID you know like polygon Nightfall so it's just uh two super Nets at some point in time so it was just a very wide product suite and polygon of course is a big player in the in this space in general but like any organization um that has to handle these many products will not be able to do justice to each product uh on its you know like stand alone right like so it was becoming difficult to juggle these two kind of narratives within the same organization in the sense that the more recent ZK evm or polygon 2.0 initiative is more ethereum L2 specific right like so so that so polygon is going strong on

that brand right like ethereum being an leading ethereum layer to player right like I mean ZK rollups on ethereum and that was uh not fitting in well with what a really wants to do from a layer one perspective so the identities are pretty different the product is very different the positioning is very different and so it's very difficult to able to manage this within the same organization and so we thought it was best to kind of uh spin out uh and uh uh do this separately uh it also allows us to you know kind of allocate resources accordingly so I would say it was more a um resource alignment uh move rather than I would say um like a rift in the

literal sense okay I got it yeah let's talk a little bit about you know the the the the the different types of security layers or da layers that applications can use so you know there is this data sampling da uh narrative that I think Celestia has been sort of carrying and Avail definitely Falls within that category uh then we have like reaking like igen Lair and um I I think we could also put uh Cosmos interchange security uh in that um in that bucket and then we have uh ethereum now ethereum can can do data available ability but it can also

do like on onchain transactions and sort of onchain contracts you how do you think how should developers be thinking about the different tradeoffs uh when choosing their data availability and to some exent you know consensus layer uh for their application right if you're an application developer you know like you're looking at whether you're going to be building on the evm or Salon ofm or Cosmos SDK or you know any number of like different execution environments when it comes to choosing the security layer like what's the biggest thing that people should be considering one are the tradeoffs that they should be thinking about Etc MH no I think that's a great question in general uh so I'll try to break this down into um couple of um segments in

the sense from a developer point of view right like I mean there are two two developers that we can talk about here so these are of course there are application developers out there but there are also these more um you know like rooll up developers um out there or chain developers as we might call them right like and so so the choice of data availability firstly goes to the chain developer or the rollup developer U uh it you know the application developer comes later in terms of because they have a set of different trade offs to uh make so I just start from the rollup developer first and then go to the application Developers for the rollup developer or for the chain developer right like I mean if you're spinning up a new new chain right like what do you

today what do you have to do right like uh so let's say you're a new um Team or developer who want to spin up a let's say a cosmos chain for example or any sort of chain right like I mean what are the steps that they they need to do right like they need to of course spin up the software using something like let's say Cosmos hdk or like let's say like get for example or even like bu hdk or you know something like that some framework uh and then in the case of Cosmos or like a independent L1 they need to bootst the set of validators for example and so on right like they have to do a token so that you know they have to do a community ensure that the token does well ensure validators are incentivized and so on right like so

there's this whole cost of creating a chain in itself right like and that um that gets kind of uh solved in a way uh by way of you know like shared security uh in terms of the roll of Paradigm so now let's say like a base uh for example right uh can be spun spun up by coinbase for example pretty rapidly without getting into this whole validator bootstrapping problem right like if you look at Bas they took the op stack and then just deployed um it as a rollup uh on ethereum and essentially the the entire work

around bootstrapping the Valor set ensuring that the chain inherit security for example um that goes away and chain develop can now really focus on GTM uh side of activities right like so the B so Bas is what what B really got from the op stack and being a rollup is that they didn't have to invent or create their own security they just inhered the security from the base layer and now they can do stuff like frch you know like basically GTM activities right like so so I would say from a chain developer point of view the role of shift to rollups is is the main thing and um from that perspective right like um if you want to do a rollup you have it's a

question of shared security Now what kind what is the level of security that you want to inherit for example uh and that will depend upon you know like a lot of trade-offs between cost uh um you know the user experience and such for example right like so if you are maybe like um like a stable coin defi sort of uh application that you know like really needs um you know today ethereum security uh for example you would you know like do a uh roll up on ethereum for example but as we are seeing with players like um you know like arbitrum and starkware for example which employ Solutions like arbitr many trust or starkware validium for example all of those are you know

like choosing uh data liity layers which are different from ethereum right like So currently those might be dac's but you know like once like Solutions like aail and Celestia come into production U you know you'll increasingly see the usage of these layers so just to quickly summarize right like um for chain developers this is a very credible alternative to kind of building a sort of independent chain um again it's a it's a complex Nuance question right like I mean um do you want to control your cryptoeconomic security or do you want to get into the shared security Paradigm right like it's a question uh and it's all right to you know like be a more Sovereign independent kind of chain

like the Cosmos chains are like so so I um you know like asking this question Cosmo was actually and you know like let's say let's say certain Cosmos chain has probably let's say 500 million inse security for example so does it make sense for them to move into a shared security context uh if they are maybe securing 200 million of tvl for example so maybe not like in that case but if it is approaching 400 million of tbll or 500 million of tvl so then then may they may want to increase security or share the security from another base layer right like that might be ethereum it could be Bitcoin uh but you know like entirely something else like a whale for

example once a whale also you know like gets that kind of cryptoeconomic security so it's a complex Nuance questions um but you know like those those abstractions will you know like become it'll be more abstracted as we go forward right now we are at the very early infrastructure building stage of you know like this separate layers of uh blockchains and quickly you know like getting to the application developer um point of view I think for the application developer um you know things are also changing pretty rapidly with this emergence of you know like ZK systems right so there are two primary things that we have to keep in mind right like one is you know like you have

execution then you have like consensus and ordering uh U layers now uh because we have these uh newer ZK execution environments like validity checking is actually becoming very very powerful in the sense that you know like even if you have like a single prer and that can thr a mathematical proof you can prove execution validity now so basically in that context what what really matters then is you know like the ordering or consensus of the folk choice on the Bas layer and so it's what basically what is what is happening is it's it's now becoming much and much easier from a from this perspective to

actually come up with arbitrary execution environments right like so it's not only like a evm environment or you know like an ABCA environment we are seeing you know like move environments for example um you know like newer custom environments like uh let's say po there's a team called poly Lang that is building on miden VM for example that has brought out a new kind of um experience uh people are taking the Mina VM and you know like doing a uh roll up runtime on top of it for example Sovereign Labs is doing a new kind of execution environment using risk zero for example so you have this newer kind of execution environments and and I'm seeing a definite um focus on more

application specific uh um uh rooll up um exhibtion environments and I think that will become um much much uh important in the future right like I mean not everything really want has to be on a shared chain like not everything needs to be in the evm for example and so for application developers as long once the correct abstractions is in place it it'll be more of choosing the right tool for the right activity and then because you'll have access to other applications using some sort of interoperability and so on right like so essentially it'll be at some point in time it will be more like uh um you know like write your program and with with

the emergence of ZK it will get proven for example and then it's just about consensus and ordering yeah so what what about um like you were talking we said earlier that you you thought that app chains uh were making their way into ethereum and that essentially the the app chain model um I mean I I kind of you know factiously said threatening but that the app chain model uh was in some some ways for some applications probably better than using the VM uh it so it it feels to me like basically app chains wanted to op like app chains promised uh execution sovereignty and uh and security

sovereignty and like sovereignty over governance and things like that but it turns out that the the thing that uh people really want or that the thing that is um better for applications is execution sovereignty in that in a DA world uh security sovereignty matters less and governance sovereignty I mean you can still have governance sovereignty with a with an application chain so do you think that like the the evm rollups um of of which there are many right where you you deploy uh smart contracts on an evm do you think that those have long-term viability uh or applications new

applications would rather just spin up their own their own VM like their own Cosmos SDK roll up or substrate roll up or whatever roll up so that they have full sovereignty over execution over governance perhaps over settlement as well if they settlement rollup do you think that we're heading in that direction uh rather than just like more shared execution in the form of uh evm rollups yeah I mean I think uh so there there are many answers to this but I'll just maybe try to illustrate this with an example right like I mean from a application developer experience point of view right like so so when you are building an application for the evm um and you know at my U time with polygon P for example interacted with like maybe

hundreds and thousands of developers right like and like like this there was this you know like let's say um someone like like a major um Enterprise wanted to um add nfts to their uh app um at that point in time for example and so if you kind of try to understand the development methodology that they have to follow right like so if you want to add a simple thing like an nft uh to existing app right like I mean um they have to run an uh evm indexer uh which is not that that easy to set up right like either you write your own indexer or you know like take up a subscription

from a hosted one for example right like and then you have to manually pass your logs because the evm is a shared State model and you have to you know like map your data model to the EVMS data model and then pass the logs like listen to every blog pass the blogs logs for example and so on for your own app data um you know like you have to run things like um you know like hosted full nodes like you have to have infer Alchemy for example subscription some of these can be pretty expensive at some point uh some point uh you also have to run transaction relay infrastructure like let's say bonomi or something like that right like just to ensure your transaction gets confirmed for example or retry or those kind of things right like so what I'm

trying to illustrate is uh it it doesn't um seem like it but there's actually a lot of infrastructure overhead that a normal application developer has to go through just to run something as simple as an nft um application right like and and that has been the reference for a lot of application developers uh that we end up doing all this infra over at time and so it's again it just takes away bandwidth from the most important part which is the business logic uh of the application or the creative part of the application right like um and so so so I mean of course there are other points but you know like even just this point you know like of course you get access to let's

say synchronous composability and so on right like but I would argue that a lot of this thing is possible in a more application or execution sovereignty model with you know like interoperability between these environment right like um and so I would certainly say the overhead uh in terms of uh creating these more app rollups for example will actually tend much much lower over time right like I mean today you know like doing a cosmo SDK chain still is a lot of work doing an applicant specific rollup is still you know like very much early but I think in the next 6 to 12 months you will see that you know like you'll have Frameworks where spinning up an achain

is like you know like a one click thing where all the infra is taken care of and you no longer have these um you know like these kind of problems right like I mean getting your transaction confirmed or you know like you like doing a shared RPC or doing this big uh indexing thing uh because you know like so it's all of this become easier in a more execution sovereign kind of you know like uh world right like and so I think I think you put it very rightly right there's ex sovereignty thing there security sovereignty thing the security sovereignty thing has proved to be notoriously difficult to bootst and maintain uh and so I think the main benefits of that is on the execution

sovereignty side and this move towards a prols specifically uh tends on the executive sovereignty side yeah you know I I tend to agree with that I I think uh like for for like Dimension is one one example right of a team that's building this kind of like roll up as a service or enabling uh easier roll like making it easier for rollups to be deployed uh to to underlying security layers so like I think that's a step in the right direction and I mean it look it makes sense if you if you compare it to the the um you know like App application Paradigm in in web 2 we moved away from like shared um shared hosting or like shared

execution environments a long time ago and now everything is a everything is a container everything is contained and execution is um is is is Sovereign and from the point of view of the application to the point now where even like functions are are are now their own execution environments right so like uh things have modularized in that way and I I think that the the evm model uh doesn't really align with that now the thing I mean like the evm has like tons of developers uh tons of um of network effect and like people who build on the evm I think have I mean for for people that I talk to who build on the evm uh I think they they feel like evm dominance is here to stay but I I don't know how that aligns with with just you know the sort

of like natural progression of how applications are modularizing and how how they did in web 2 and I think like how it's inevitable in in web 3 so but yeah I want to get into the to Avail a bit more uh and talking about uh the architecture and um the different components of the stack so um yeah what are the different components of aail and like different components of the stack and one question that that came up when I was asking on Twitter uh for for question is uh why you chose to build on substrate and um yeah what was the the motivation there and then we can get into other things like consensus and the proving system and data sampling and all that yeah yeah well makes sense so quick I just I'll give you like an overview of the a architecture and then we I can get

to the substrate question specifically so the architecture itself is that you know like actually that we we build or maintain two networks actually so there is this Bas layer validator uh run network um which we call the a Bas layer and then there is a separate light client datability sampling light client Network which itself is a network in itself in a because you know like all the participants have a P2P um layer in between and then the lightline network talks to the a Bas layer in in in that sense right like so so essentially you know like a wh is comprised of these two different networks which are connected to each

other the base layer itself uh is uh built off substrate uh so we uh when we were starting out we had this um choice to you know like what framework to base our development on and so of course the choices were tendent and substrate um uh at that point in time right like and we had previously uh used tenament at polygon POS while building polygon POS and so we had extensive experience using um you know like tend for there and um substrate was uh was maybe a couple of years into development at that point

in time uh and you know like like when we started looking into it in 2020 it was surprisingly uh you know like well done well engineered uh modular uh in in the in in the you know like in the real sense in the it's very easy to actually switch off or change some functionality without you know like disturbing the whole code base for example yeah I think they were very early to to like building a very modular modular development stack like even like in 2019 and or even earlier they were quite early to building like this very um modular stack yeah yeah yeah yeah um it it's also you know like uh the the you know like our

Tex stack is primarily rust so it was like great that you know like it was written in Rust for example uh um and also you know like from um from what we got out of the box from uh substate was also you know like uh grandpine Bape consensus uh this is like not maybe well known outside of the P ecosystem but you know it's it's like really um you know like well written in that sense Grandpa B consens is like a hybrid Ledger uh which you know babe does block production and grandpa is responsible of finality so it's like a hybrid uh Ledger where uh you kind of maintain block production via babe and then there's

this finality gadget on top from Grandpa and so um it's possible you know like for the chain to keep running and then finality can be reached later in extreme circumstances mostly finality is reached within couple of blocks but um uh in general in terms of network partnering and those kind of you know like edge cases uh you can always do a delay finality and so it's basically like um I would say in the direction of you know like things like Gasper on ethereum for example in that sense so it's slightly different from let's say tender uh which has this instant finaly thing so so

different characteristics right like I mean tender will halt if if there's a you know like a uh a problem there in in that sense so I mean of course there are pros and cons etc for example then I think we also were we we are able to onboard like um support about up to thousand validators with uh Grandpa consensus um and uh that was attractive to us because we will be able to support a bigger Valor set um uh then let's say on on tendin for example and one of the very underrated things from substrate is this um npos or nominated proof of stake system which essentially you know like

so it employs this implementation of something called fragment uh which ensures that you have stake distribution uh you know when people nominate or delegate their tokens it doesn't go to a single validator unlike in uh you know like tend or Cosmos POS for example right like so um it it kind of ensures in protocol that the stake is roughly the same across the entire valetor set so it ensures that you know like we have good stake distribution mechanics and this was actually a big problem for us while we were building polygon POS like even today it's a problem and we try to address it with a variety

of mechanisms not necessarily in protocol but also of protocol but the reality is that on polygon POS and in most Cosmos chains also which use tender POS uh eight six or eight of the top validators really control more than 67% of the stake uh um which is which is a which is a problem that we would rather avoid um uh so because ESS essentially what we envision is aail to be like the base layer for a lot of rollups on top and we want this to be really decentralized and so we started with the foundational Tunes required to make that happen um essentially so I mean it's a

combination of all these reasons uh but yeah that's yeah I mean that that's interesting let's talk a little bit more about the consensus I think this is like I I don't like to I don't want to like compare aail to Celestia so much but I think like one of the main differences at least and I think there's two kind of main differences if you kind of like look at both protocols side by side is one is the proving system uh where aail uses validity proofs while Celestia uses fraud proofs and then the other is the consensus and um I think this is sort of similar to the way Cosmos and ethereum are are different in consensus ethereum can scale to you know many hundreds or even thousands of developers uh or to validators but but um uh have

probabilistic finality now that's a trade-off that ethereum makes because it wants to be fully decentral as decentralized as possible with as many validators as possible and and Cosmos has this in instant finality but uh tournament consensus does not allow basically it starts to fail like once you get into the you know close to 200 validators or so and and I think this is one of the main sort of philosophical differences between AA and Celestia where Celestia's built on tender and and has this constraint but instant finality and Avail um uses this um this probabilistic um uh consensus mechanism I I'm curious what uh impact that has on uh the the data availability uh

sampling uh aspect of basically what you guys are building does it have any implications at all or or is it only just block production that um is is is impacted so I mean uh let me take it from two angles right like so so one is of course the consensus part the second is of the like the other uh like um let's say the implementation differences between us and Celestia right like the validity proofs and fr proof uh differences right like so um in general um um if you kind of uh take a look at the uh the validity proof and fra proof based uh differences right like so

once you once the block is final on the base layer um the headers are propagated to the light clamps for example and then you know like this whole random data sampling uh process uh flows and essentially um uh that really the Das verification really depends on whether on the type of proving system that you have right like so in a validity proof system you can actually verify it instantly uh on FR prop system you have to wait in like a certain time interval it might be the minutes or hours for example uh so uh

from from a application uh or a developer experience angle right like I mean think of like a ZK rollup uh running on top of um uh a whale uh and uh Z running on Celestia for example right like so so it the the thing is that the for finality you have to wait for two things right like one is the da finality and then the proof let's say the proof generation of the execution proof generation right like now it might happen that the execution proof has been generated but you're still waiting for the da fra proof to be propagated for example right like so time to finality in that sense increases um uh in let's say this the same

construction with the ZK roll up on a whale right like I mean as because our construction really uses validity proofs right it's as as soon as the block is finalized um and of course we have deterministic finality but in most of the cases uh uh finality happens within couple of blocks um um you know like so essentially in the majority of use cases uh you'll be able to kind of get da falty um and like in the case of ZK up like you don't have to wait around for that right like so those are the more philosophical differences and to zoom out a bit on on the other side right like I mean um it's primarily about what is

this uh role of of a datability layer or a base layer as we see it right like so today if you see ethereum is the most dominant data availability player in that sense from a model perspective of course all blockchains have a datability component even Co like Cosmos blockchain Solana Etc have it embedded Within in their um uh stack but like from a modular perspective ethereum is actually the most dominant data availability provider to rollups on top right like that's the thing today and if you really want to kind of provide a compelling alternative to ethereum in that respect right like you also need to be a very

decentralized um um you know like secure uh layer right like and so if you want to really get into that um uh um that that kind of positioning like you have to think about these things right like it because it's very easy to get into a um situation where you know like in in in especially using tender that you know like eight of the top 10 validators or eight or 10 of the top validators really control uh 67% of the stake right like and so for a bige layer to have that kind of mechan it's a it's it's it's risky dangerous you know like it's not the best situ situation right like and

that's why we consciously took this approach to kind of um be closer to the ethereum philosophy rather than the cosmos philosophy right like the cosmos philosophy in in at least in my opinion makes sense in a more application specific chain um uh uh perspective but if you are really trying to provide a more shared security kind of product then you know like the ethereum philosophy matters more in that sense um see having said that there are of course pros and cons right like ethereum has like this ability to on board like hundreds and thousands of hundreds of thousands of like 100 thousands or like

more maybe right like but the the problem is that you know like there is a limit to communication complexity that a network can manage right again so now there are discussions within the ethereum community also to to increase the 32 e um you know like deposit uh limit for sticking to I think 248 or something like that so because they want to ensure that because see the reality of what is happening is people want to St the people who want to stick a lot of eat really end up running redundant Val nodes um yeah and and so essentially increasing the traffic uh on the P2P traffic on the network right like and so the there is a proposal which is under

active discussion to increase the deposit limit to 2048 each so that it just reduces The Valor number and so so I think the answer is somewhere in the middle right like I mean p uh you know made a lot of good choices um on some of this right like I mean it's not as low as let's say you know like a 30 or 100 kind of a figure uh although I won't say 100 is is a small figure like it's a decent enough figure but you know like you don't want to be in the hundreds and thousands and then kind of try to scale back the number in that sense right like so we thought thousand was a good enough number and on top of it if you kind of look at it from a npos kind of a um uh like uh

implementation point of view like you know how many unique validators are really running your network right like that is also an important metric you know like people like to call it the nakoto coefficient and so on but even that is important right like you can't have the same Val is running multi too many multiple nodes right like so what are the lers that you can employ uh and so so this is like a broader question right like I mean eum is grappling with let's say things like like Lio like um uh and you know like ethereum reaking and you know like those sort of things right like so uh so at some point in time we will also you know like get into those kind of questions and we really wanted to start off with

the right foundational levers U to kind of you know like solve this right like otherwise you know like changing some saying production Network in between like changing the things like the consensus is like a very it's a very big move for a network right like um and um so yeah hopefully I was able to answer some of your questions yeah I know that that that that that all makes sense and I I think uh really clears clears that up so but yeah let's talk about the data availability sampling then um maybe you we've talked about da sampling on the podcast before but uh maybe let's just unpack that and remind our listeners like what that means and how

applications use it to in order to uh prove that their data has been essentially um uh added to a block without having to add the entirety of the data cuz that's one of the the things that um platforms like Avil are that's where they're different from like ethereum for instance where like all the transaction data is on chain uh Avail only sorts uh stores the the the the the hashes of the data um so yeah maybe um explain da sampling yeah yeah uh no so so let me just also correct you a bit right like so uh in general AA stor even even I got it wrong so we need an explanation yeah

yeah so I mean see I think just zooming out a little bit from your question to answer this right like so essentially you know like what what really goes on in a blockchain is that you know like what you really need to know is is the correct order of the transactions um uh um and basically the folk Choice Rule and then you need um uh you know like the execution validity so in a in a conventional blockchain Network let's say like a cosmos um uh blockchain right like so uh there might there might be like 100 validators there might be like few hundred uh noes full noes for example and so on right like so what the validators are really doing is you know like they're getting the transactions uh

and they're you know like let's say in Cosmos for example there's a block Builder which will kind of order those transactions into a block and then each validator will reexecute those transactions and then uh the 2 by3 of the validators or 2 by3 plus of the validators will reach an um understanding that whether on the folk Choice rule right like primarily like which which is the correct block or the correct order of transactions um and then the full nodes will you know like listen to this uh they will not trust the validators um uh uh with with the validity execution validity they will download all the transactions uh the and then re-execute

those transactions and then they will arrive at the uh right state for example right like so that's the rough uh very rough uh flow in a conventional Network so now if you want want to put this in a more model context right like what do you really want to have right like so in ethereum context what is happening is you know like there's this roll up on ethereum which exe takes in transactions from users um and it's called a sequencer uh today because that's the component which orders those transactions like which defines the order of those transactions so and then they commit to the order of transactions on ethereum they put the whole transaction data or the state diffs data on ethereum so that even if

the sequence are now goes down right like anyone can take the state from ethereum and um you know arrive at the correct State because they can just reexecute those transactions and now so the different rollups differ in in you know like how you can verify the execution right like so there are probably three types of these different transactions right like pessimistic optimistic ZK so pesimistic is again you know like take the order and re-execute those transactions again optimistic it is that some people will uh look you know like re ex transactions and if there's a problem then they will raise a fraud proof for example you know like a ZK roll up or validity proof thing is that there's a prover that provides a mathematical proof and you just have to

verify the proof right like so now like just to like give the entire picture right like there is this whole ordering piece and then there is this execution piece now naely if you want to look at it you will download all the transactions um and um reexecute all the transactions that what that's what happens in a conventional blockchain Network now the question is how do we make this process more efficient uh for users or you know like just in general stakeholders now in ethereum for example today you will have to download all the transactions to get a full datability guarantee um um and if this happens in any P Style Network right like the main

Innovation that aail uh brings to the table is efficient way to do this um data availability verification like you don't have to download the entire data the entire data is there on aale uh for sure right like aale certainly stores all the data in the ordered format now but you don't really need to run a full note and download all the data and then verify data ability this data sampling thing it employs this clever uh mathematical trick of erasor coding and kg polinomial commitments in our case uh to enable data sampling where essentially you are kind of so this is like just to simplify this right like

eraser coding is this technology that is used in has been historically used in you know things like storage like you know like on a compact disk for example um if you if if your listeners remember compact discs uh if is that you know like you can you can have a compact disc or a hard dis for example and you know like maybe there are some scratches on the compact dis uh and but you want to ensure that no data is lost so what compact dis manufacturers do is eraser code the data on the compact disc so that it is replicated and in randomly and such a way that you know even if there's a scratch on the compact disc even if one data fragment is lost you are you know like it's very easy to

ensure that the uh a replicated fragment is available somewhere else on the compact dis right like that's eraser coding roughly right like and so essentially what what this allows is for these light LS to verify this data ability very sust like within a few samples they're able to kind of verify up to 99.99% uh probability that the data is available right like and so that's one part of it the other part of it comes from the more execution focused uh uh uh verification which is like let's say let's say ZK proof ZK rollup right like essentially you don't have to reexecute all the transactions right you just verify the proof of the transactions so what what I'm trying to arrive at it is

is that this unique combination of Das which gives us very efficient da verification and more exe sustain execution verification so the ZK rollups are the most easiest to verify but optimistic rollups you know like are not far behind right like I mean the sense that not everyone wants to do it someone will do it and you optimistically know that the execution is correct right like so again pros and cons in both these uh but so essentially this is what I'm trying to kind of convey that you know like you have to look at it from an entire angle and that's what really datability sampling it it kind of um brings the ease of datability verification and when you combin it with

execution verification it becomes very very powerful and that's why we always talk about aail more like a like a enabler for these newer kind of uh rollups for example of course it will support existing kind of rollup constructions uh but you know like you can do entirely new kind of rollups with this kind of architecture again last Point really quickly I just wanted to cover there was because we've been building this data availability sampling uh technology right like so um what this really are the are these are these light clients um which do you know like these random sampling and we' have kind of used something called cadamia DHD um to implement this on on on the lightline layer and

essentially you know you can like just to simplify things um this is you can think of this like a bit torrent kind of um scheme it's not the same but you know to visualize it's a useful analogy in the sense that you kind of put the data on to ail the transaction data from rops on to ail after gets finalized it it slowly gets randomly sampled By the Light CLS and also fragments of those the data also get stored on the light CLS on the cademia dstd the DST is a distributed has table and so any new light Cent that comes into this network doesn't really need to rely on a full node or the validators to get this data is actually available of um you know

like a light of the of a light Cent that has already uh you know like uh sampled some data for example and so by the dent of this what we had to really build was a very comprehensive P2P Network um actually to kind of enable that you know like stuff doesn't get overloaded and you know it's pretty efficient and that um will also serve as a foundational layer for this more execution uh verifier uh propagation right like I mean in the sense that you can actually now take the da da ver like line which is a verifier for da you can take the execution verifiers which is let's say

like a recursive proof verifier and then embed this into wallets for example right like and so the combination of this is extremely powerful again sorry long waited answer but I just wanted to ensure that you know like I cover all the angles yeah no that uh that's really interesting and so um like what what are the like if I was reading on your on the on the blog that you you guys are like there's this interesting research about increasing the block size so I think like right now the the the the uh eraser coding is like 256 rows by 256 columns and there's some

particular trade-offs when when you want to increase the block size uh can can you talk about like what the trade-offs are and um what is that basically dependent on are we depending on like uh better uh better network connections in order for like blocks to propagate faster is it more so like being able to validate those blocks so when validators run the transactions or is it the Eraser coding that then takes longer like what are we what are we kind of dealing with in terms of scaling here well I think I think uh that's a good question to ask in that sense so like what changes when we actually see why are we again building a whale as a separate like a layer that focuses on one aspect the

data ability aspect or the cons ordering aspect right like I mean why are we focusing on that because when you decouple execution from uh let's say the data availability or the consensus and ordering layer what now happens uh is that we don't really have this uh you know like issue of maintaining State on on on a whale for example right we don't have to maintain execution State uh and that really allows us to increase block sizes uh pretty rapidly uh um so the alternate approach to this is like let's say like a more sharding sort of an approach for example but really in this kind of construction we don't really need to Shard right like we can just increase the block size and it's very

different from increasing block sizes on let's say uh on execution Focus blockchains like we had this whole block size scaling debates and Wars on execution Focus blockchains right like and it is it is not practical there because you just increase the requirements of you know like let's say full nod and well nod and such in this specialized um you know like construction which aail employees right like we we can increase block sizes pretty rapidly like we have Benchmark sizes of up to 128 MB uh uh and it can go even Beyond right like I mean because we have this very neat construction of eraser coding and polinomial commitment Generation Um now okay what happens when we actually increase the block uh block

size right like in that sense is that um you know of course the time to commit uh generate the commitments on our blocks we have this polinomial commitments that we create and we add it to the header so you know like mathematically you can think of this like a m cross n Matrix we eras a code The Matrix to make it 2 m cross n and then we do commitments for the row for example so so if you increase the size of the block we are really increasing the size of the Matrix uh and so the number of rows increases for for example and so the commitments that we have to generate also increase for example uh and so on right like but essentially like the the polom commitment process or or the modules

that we use right like these are not really um you know like so you can't compare this to something like an evm ZK evm Prov for example you just just very simple uh circuits that kind of generate the poal commitment right like and these are pretty we have been you know kind of working on iterations and these have become more performant over time right like so so for example for a 2 MB block these commitment uh is are generated within milliseconds uh for 128 block MB blocks these will take a couple of seconds uh for example but our because we have this you know kind of Bas layer we don't really U handle execution we don't really need to have like extra

block times right like uh because the extra fast block times are on the execution side on the rollup side on the sequencer side so that's why we have the uh capacity to actually increase the block size even without increasing you know like keeping within the constraints of the block time that we have set for ourselves right like um so so those are some of the trade-offs that we have to make um in general because we stick to this validity proof approach uh and we generate these commitments with within within like very uh quickly uh we can propagate this um you know like proofs can be generated uh for these commitments pretty rapidly and then you know like they can be verified pretty quickly uh so that's the beauty of

having this different sort of data avilability focused layer so to speak I may be missing some of your questions um but hopefully I've answered some of them but just feel free to remind me of yeah no that that that that covers it um so I want to get into the the economics of aail a little bit so like for for an application you know when thinking about the monolithic stack when you're building an app on ethereum like your cost structure is is pretty easy to to sort of reason about right you have to pay ethereum for Block space um or users of your application pay ethereum for Block space unless you're subsidizing that somehow in in the modular stack um you know depending on the say

the rollup as a service that you're using maybe you don't have full sovereignty over the the the sequencers as a settlement layer uh and certainly the da layer you sort of like choosing an underlying stack based on your kind of execution layer provider um you know how how do developers have to reason about this and how will because essentially like I think the competition will be on on price it it like competition will come down to like all of the layers of the stack constructed together in like um in in um uh in a stack right so you know if you're building an application that is

on a rollup as a service provider and that service provider is using Celestia or ethereum or or a veil you're you're sort of beholden to that whole stack and and their cost structure in order to figure out what the cost of your application will be is that the right way to think about it and uh and then what do you anticipate the cost of running application will be firsts compared to you know monolithic l1's today do you think that the module is that will reduce the cost of running applications uh overall yeah yeah no I think again good good questions to ask right like so I think you are certainly right like that cost will be a factor at least I would anticipate in the initial short to medium term is is what we

anticipate right that's how we are looking at it uh from that point of view uh certainly cost will be the driving Factor initially um we expect that this will be augmented with you know like uh us enabling newer sort of Ro of constructions for example uh where you know like you have um uh you know settlement happening Clans side for example and like host of newer um rollup constructions that can happen right like more like let say Sovereign rollups and such right like but we'll not get there the moment right like essentially initially you know like I think you know like we I kind of like to use this analogy of you know like cars and wheels uh for now that you know like uh like

rollups are really like the cars um car manufacturers and we are like the wheel manufacturers right we are like the oems on the uh on the cars right like and so essentially our GTM primarily is for you know like to to to you know like um partner with let's say cars to ensure that you know like you know their cars go out with our wheels for example right like and so essentially um um the cost structure also kind of you know like reflects that in in that sense right like so um so in general of course the sum of the costs of each layer will be exposed to the application developer um today because there is um you know like let's say a

less da capacity on let's say something like ethereum costs are pretty high uh ethereum price itself plays a big role also uh so in general we did some modeling uh on the cost structures and like over the last few months the average price of um uh da on ethereum is about $13 per MB uh $1300 per MB uh of data um this is based on actual $1,300 per megabyte yes $1,300 megabyte roughly uh of course there's you know spikes in E price and usage and such uh what

whatever it's paid but the rough price is around $1,300 per MB of data right like I mean um essentially the trick for us is to because we have this whole increasing data like scaling data capability because of you know like our construction we are able to kind of you know like of course uh bring cost down because we have more da capacity right that is certainly and uh um but we also have to ensure that we are sustainable our validators have to be uh profitable for example we need to have a wide enough validator set for example so we have to be essentially we we will be more costly than let's say like a data abity committee for example which is

like you know like run by five or six seven data members right like so initially there there is this whole cost game that that will come into picture which will you know like kind of uh play into the actual application developer cost right like so you can see right now let's say let's say like an arbitrum runs on one chain it also runs on NOA chain uh for example uh starkware is planning a volution mode whe it you know you can run in rollup mode or in validum mode right like uh all of these uh the NOA solution and the validum solution right now employer D for example right like so so the cars or the l2s for example have really literally already

made the choice to you know like kind of use layers other than ethereum uh for some of their offerings right like um and so essentially uh what what we think is you know of course we will be um from a cost perspective comparative uh very very competitive against let's say like a uh ethereum but like maybe more than a DC for example in that sense uh because we offer different characteristics of decentralization and security and such uh for example um if you want to compare it to like say like monolithic L1 chains right like so what really differentiates is is it just in general the

sheer variety of exhibtion environments that are really going to come in the next 6 to 12 months right like I mean because with this we are able to kind of support a variety of exhibtion environments uh with which you know like bring a different sort of trade-offs for application developers uh so so it's a combination of this right like the costs we will have to of course play as per the market uh how the market goes we have to ensure our validators are profitable uh and then we have to ensure that you know like kind of applicant developers get a good deal so it's it's somewhere in you know like that takes care of all this um hopefully that answers your question uh yeah that's interesting I that I hadn't heard that

figure of $1,300 per megabyte it seems like so incredibly High I mean if I know I know this doesn't really compare but if you compare it to something like R weave I think the cost per gigabyte is like less than $100 or something right it's like $50 a gigabyte for what you know you know eternity apparently um so yeah I don't like yeah I mean see basically of course that uh same metric like the data data is very different like let's say like what an RV and a file coin really store right like I mean so here the product being sold is you know like ordering consensus and ordering like data availability is

really not data storage in that sense right like it's yeah yeah it's it's basically a you know like uh which uh which is the correct ordered set of transactions and so so technically it is very different from you know like a RV of five coin and this is the only reason why um you know like uh so ethereum is the only uh I mean not not the only player but U because each L1 maintains its own data availability and so there's a cost to that as well for example uh but it's the only player that has you know like where rollups are there right now and it is providing da separately uh at cost like at some cost to uh you know like

customers on top right like and so that's the reference that we have and uh and $300 is right now where e is um you know like at not that uh higher price for now right like I mean uh uh it it certainly goes higher um you know like if you get to data from last year for example or the year before right like so so essentially uh certainly um uh that's see that that's the main reason right like I mean see there is a price for for this product um you know like and the current um so see I don't really want to directly compare these Solutions but

what I'm trying to say is that the that ethereum certainly charges a very high fee at the moment with its upgrades it will come lower but there is still a wide uh range uh of cost that you can actually maintain and still be profitable right like and so so there's enough space for you know like like a for example and even celesia for example to come and you know like operate at um you know like in this market and right now very frankly there are not too many players uh in the market uh yeah yeah cool well I um I'd love to ask you more questions like I've got so many more questions here but I know we're running a little long here

but I I I want to ask you like one final um question about basically what what's the state of Avail currently I know you guys have a test net uh what's the road map and some like major Milestones that people should expect and um yeah maybe ending on like one of the biggest challenges that the project still faces before um mainnet yeah yeah so essentially um I would say um so we have slightly um in know different um uh we different kind of company in the sense that you know like we we were start initiated within polygon and so we have been actually been in development for close to like two and a half three years now and so we

are actually very near Main net um so uh in the sense that our second test net is you know like now concluding our final incentivized test net is um starting in the end of October or first week of November um it'll probably run for a couple of months I think it's Christmas season so things will be slow over the holidays but I think January is when we'll conclude that and our final audits are finishing uh pretty pretty soon uh by the end of the year and hopefully you know like uh we are pretty much code code complete right like from a minute uh release point of you so we are just waiting for things like incentivate for

issues to occur uh to be to come up there for example or from the final audits for example but otherwise you know like we are we are very we we kind of are ready for like a q1 2024 minute uh launch um uh so from a development stage you know like we are actually very near production um our Bridge from a to ethereum is also uh expected to finish by December for example we already have optimistic Bridge uh we are working on a ZK Bridge with the the suin lab folks on that um and that is also like a major Innovation there um in General we've been busy building Integrations to Major

rollup providers uh we we are going to introduce um you know at least three major um uh ethereum L2 roll up stack kind of Integrations pretty soon others will hopefully follow pretty soon so you know like I mean it's now we are very much getting nearer to uh like lot of rollups running on top of you know like test net initially then on Main net as well we you know like someone like a dimension like we Dimension works pretty closely with us they running incentivized test net and if you kind of look at their portal there are literally hundreds of rollups running on the uh you know like a test net uh even uh like as we speak for example and uh that's

cool yeah and uh so so I mean I think um we are pretty close to minute um so so yeah so we will'll be announcing some of this stuff pretty soon um in terms of uh like problems you know like um I think uh from a I think from a technical angle uh solution if I if I will you know like I might say might say like aale is one of the one of the best taxs uh out there in terms of you know like the architecture the implementation for example uh but uh in general from a from a marketing angle I will really uh

readily admit that you know like we are behind the curve in terms of let's say a Celestia for example which has done a great job of you know like kind of creating this whole data availability or model category and so we are kind of you know like in that follower uh positioning at the moment right like uh and I will readily admit to that so that is I think the main problem with ail at the moment right like uh uh because I would say a validity proof Bas approach um uh to solving da is is the uh is the better approach I would say uh because you know like even if you look at ethereum they are really following the same sort of approach U they want to do it uh uh and so we are really leading uh

in implementation from that sense uh but yeah if you ask me one problem that we have this is that I would say that is the problem and of course we are looking to kind of solve it and hopefully you know like uh we'll get to that point but yeah I mean uh we'll take it one by one we just want to kind of get to Main it U soon and then you know like slowly we will U solve those problems as well yeah well maybe uh what you lock in marketing uh you make up in uh branding and and like technical writing uh you know the the Avail blog is like really really good and there's really excellent articles

there I'll I'll link to it in the show notes and I I'd encourage anybody to uh uh to that's interested in Avail to go check that out read um some of the Articles you have there and um yeah this has been really fascinating I've learned a lot about Avil and da and really looking forward to seeing uh Avail go to mainnet and um and yeah compete compete in the space because there's there's all sorts of interesting uh projects building in the space and I think the the modular stack is is definitely not like codified you know it's it's still very much in flux uh and I expect that to to happen in the next maybe 3 to 5 years right where we

get a much better idea of like what the stack is looking like and we didn't even talk about interoperability which I think is a big component here but we'll have to save that for another another time an thank you so much for making the time today and coming on the podcast and sharing all these great insights with us yep yep yeah thanks for inviting me and you know like it was a great discussion thank you