Data Availability Summit 🗄 Avail: Architecture and Use Cases

[Music] yeah let's uh talk about aail and how the architecture looks like and what are the kind of use chases that it enables so uh just before we go into the agenda about what are the things that we'll be covering I'm plal I am a Kong uh along with unrug unrug was previous year profound of polygon uh um earlier this year we took AA out of polygon as a spinoff and uh to create a separate entity we want to create oural base layer for chains and in this talk we will mainly cover you know how blockchains are changing uh what is it that we are building how does it look like how are we go how are we trying to solve the problem of data availability uh what the base layer constitutes of how does the light plan Network look like and then what all of this means what does it enable what are the kind of use cases that it enables uh where are we are at in terms of development on the timelines and the future optimizations in terms of how we can improve and so on and so forth so yeah I think uh till now you guys have already heard a lot about how um in typical the uh rollups are like taking the center stage and we have already recognized them being the main way to scale computation and what a rooll up is is basically it acts as a valuating bridge which primarily allows people to do offchain execution uh with etherum assets so that's how the rollups in the ethereum ecosystem are kind of stealing the ethereum um um computation in terms of blockchain construction we are inherently seeing more and more modular constructions where people are taking a single piece of this entire stack and optimizing them and making them more efficient in in ways and means in which they can speak to each other communicate with one another and be more efficient at the particular problem they are trying to solve and the one of the main carats of this entire construction is that they are only talk about who will scale the computation and where how the execution scaling works but there is another problem which needs to be solved in tandem which is how will the data scale and that is where we come come in and the rollups are hungry for data availability and we are trying to design a base layer which is optimized to solving that exact problem the main key takeaway that we have uh is that every base layer blockchain is eventually going to be a deayer we are seeing Celestia thanks to them for hosting this Summit and they are creating an L1 which is focused on da we are also seeing you know newer EIP is coming in which is where ethereum is trying to transition from uh being an execution Focus chain to being more da Focus chain and this trend is only going to be more and more as people try to bring more applications to the the blockchain space so what is aail and how does it come in and fit into this picture uh so a is a base layer it provides one simple thing which is raw block space for the rollup chain to use it makes sure that the particular data that is submitted that they are sequenced and packaged into blocks and published now what this means is that it it only makes the data available for a short period of time so that you know validity proes or fraud proes for the execution can be verified and be usable for later and how it works is that it segregates the transactional data on a per app basis so that multiple rollups can publish all of them at the same time but each of these individual applications can query only their own part of the data now how aail is bootst and secured is that it will have a robust validat set that will sign off published blocks but they do not do any execution in terms of interpretation of the data that are that is submitted to them so it is very dumb in that sense it takes only the blobs and just makes it available and each rollup which is built on top of a will has to interpret its own transactions and they can be completely custom we will talk about app specific uh rollups and such but just to give you a overall picture it can be anything from EVN to being no VN at all which is extremely custom and very s specific U and but how does work is that the light clients they leverage uh K poal commitments eraser coding and Das to allow the verification uh by downloading a very small sample of the data but not having to download the entire data by itself and what this means is that they can download all transactions for a single application rollup and they enable the application from Nots but they do not have to download other application data anymore so there can be on light CS which download only a small chunk by random sampling but there can be application full noes which download on only the application specific data and not have to download other application data but what does this mean is that it means that uh a vibrant ecosystem can built on top uh different types of rollups can sit on top of a wh they can be adapted for their unique needs anything like validum and N3 uh ecosystems can be built on top uh you can have many types of bridges I think recently many talks about what is it does it mean for one thing to be a rollup we call the rollup uh it can mean many things depending on where you're coming from this interpretations and such definitions of how you define security how you define liveness and things like that but no matter how you think about a rollup it can all of this can come come together and work on top of a will and what this also means is that State validating Bridges the strongest type of interoperability uh right now available that is also possible and this allows for asynchronous form posibility across the rollups which work on top of and because of this because that you all use the same DNA they can have shared security so applications which are building on top do not need their own value set you are already seeing like roll up as a Sur this kind of uh infrastructure where you just do one C deployments of newer chains this is all possible because they don't have to think about how to Bo stop their own security as the base layer takes care of the um and and in the base layer how does it all work so essentially it's very simple it's almost like every other blockchain how it works uh but with a Twist about how exactly the block construction looks like so when the original data comes uh they packaged into a block but instead of sending it as it is it's actually eral coded and then a commitment of the data is pushed into the header the header along with the BL block the entire block gets sent to All The validators Who sign it off and then uh they like they reach fun sensus on this particular block to ensure that it is it is it is present with all the validators and but this is not where it ends because this would have meant that this is just another blockchain with uh no uh no scalability benefits so we we will have light clients which do the sampling but I will come to that in general for the block production itself the applications send all the data blobs to a win using a particular application ID these blobs are chunk together and a particular 2D metrics is populated uh they are segmented according to this app ID the row commitments are generated erasion coding is performed on the commitments and the homomorphic property of kg it ensures that commitment of the Eraser coded data is the same as the Eraser coding of the commitment itself uh The Matrix is adjusted according to demand so and the demand is low the cost is low the block sizes are low and so on and so forth but the demand is high it kind of grows up to a particular Point um the header it C on the commitments the app index and the sizes of the MS so this is what I was trying to mention is that we have the data it's broken up into this 2D 2D Matrix they are extended column wise to make sure that there is extended data and enough redundancy we go ahead and commit to each of these each of these rows to create the commitments but the nice thing about kg is that if you only have this particular uh side of the Matrix you can just produce the commitments and extend the commitments and you do not have to extend each of these columns individually and how does the Bas layer stack look like we have built the main blockchain using substrate which is highly performant and in this consensus we use Grandpa and babe together so essentially in consensus systems there is this availability family dilemma which means that that you cannot be highly available and still have instant finality so just to kind of address that situation we have bab which is the block production engine which is has been optimized for lightess we have Grandpa which is the finality Gadget and the leader election is based on a veryable random function and the finality is not reached on blogs but on change which means that even if the finality holds for a particular amount of time when enough people come back up when 2x3 when more than 2x3 is available to find L blocks they can actually find an entire change and not have to do it blog by blog and what this essentially means is that it is tolerant against temporary Network partitions and a large number of notes fashion what this also means is that we have nominated proof of state which is a little bit different from delegated proof of stake where we see that when people come with large Stakes they inherently get larger Powers this means that they get larger rewards and that leads to State concentration which essentially means that even though we have a very powerful base layer it essentially turns out to be very lopsided in terms of power accumulation and to avoid that we we have you know something like fragment election which has proportional Justified representation in terms of nominators and nominators can choose not a single validator but but have a ranked choice of validators this allows for high high decentralization factor due to Fair distribution of rewards and that allows us to have around thousand validators in the active set um who participate in finality of the chain now coming to the light Cline network uh the light c network what it does is that we have potentially we can have potentially millions of light clients the light clients will all do very small amount of work individually but collectively they will have sampled the entire block and much more with redundancy so even if a validator goes Rogue um and tries to withhold data the light Clan network has a peer-to-peer network of its own which means that a replica of a block always remains available there and it can be reconstructed from the P top so we have built the light client from scratch then build in Rust it's it forms an overlay P top Network it samples and and uses the full node only for initial bootstrapping but on the on the back end like it uses a Cen databased DHD H which contains cell level proof so anyone can verify the correctness of the samples and with when we scale up with more and more light clients more and more uh data can be digested in as as part of a w block and there is no need to trust the validator set for the availability of the data the light clients can perform data sampling and verify availability even the super majority of the validators of a ra wants to suppress a data and that is a powerful construct right because uh that means that once enough light clients have sample the data remains available and there is no need to rely on decentralized on on a centralized RPC providers for getting any guarantees when you're using aail and that's a stark difference from how you use traditional blockchains today and not only that we are what we are doing is we are adding the support wallets where they verify both exec ution correctness and data Val guarantees to make sure that you do not have to rely on cryptoeconomic guarantees for the correctness of the data or the data availability you only have to uh rely on them for for ordering in terms of performance this was this is a bit dated last week or 10 days back we released uh an improvement where commitment timeses have been improved by around 70% so this has been reduced much further currently we have the 2mb block size uh with a few tens of millisecond for block for building commitments but um we have tested up to 128 MB and Beyond and it works so kzg and so on is is not going to be the bottleneck the bottl men is coming down to you know block propagation and such so that's what we are trying to look at right now in terms of what it enables is that it enables Sovereign rollups valm optimistic chains and so on I we talk about them in a minute but what it also enables is that you can have custom execution and custom state which means that you can deploy applications which are very very close to what you just need for your application and but on the other hand you can also opt for you know ethereum virtual machine or Solana or anything of that sort any standard virtual machine you canm and we also have you know support for let's say lups or some of the other uh execution environments as well uh in terms of validium and optimistic chains there is a DAT attestation Bridge which is live on testet what it does is that it takes the attestations from ail and pushes it to sepolia network for now but you can basically plug it in plug it into any settlement layers and then you can run a validium on top where the validum kind of just just sends the proofs or the frood proof uh or a l thereof to ethereum and the all the data availability guarantees are provided by aail through the attestation bridge to ethereum or other networks and that's how you can not only build like L2 validium on top of ethereum but you can also have L3 validium on top of Entry chains this also provides the basis for Sovereign roets what that means is that you know today you will see that most rollups rely on uh and settlement layer for validating execution groups but uh they can also be doing very well on a light climent and the L1 way of doing things is hard to upgrade or you have to have a multi and things like that social consensus is probably not possible because you need uh and L1 to 4K uh and so on but for Sovereign rollups that do verification directly on application like clients they don't have to rely on a smart contract they can upgrade by upgrading their clients just like you upgrade an L1 and the like clients can verify not only data availability but execution grps without downloading all the transal data and many different designs which were um which which are possible in this space optimistic ZK with a sequencer with a shared sequencer without a sequencer all kinds of things are all possibly in this G in terms of where we stand uh we we right now are closing up our long running test net which is a c test net where we have tested around 105 top external validators um and we are uh we have pushed perform huge performance improvements they say a week or so back next month we are going to launch our device testnet so I would welcome it everyone to kind of come and participate in that and hopefully only next year we will have online net this is our Target in terms of future optimizations where we can improve and so on right now the validators can verify commitments uh but um you know every valuator has to verify all the commitments From the Block producer because they have to regenerate commitments but we are trying to look at improvements where that is not the case we are also trying to we already have a PR but we are trying to test and include the multi strategy commitment so that you know we can generate uh not only cell level but a particular submetric level openings which will mean that there are huge improvements in terms of how the DHD is populated how the network streams work how the bottlenecks of the peer-to-peer are are coming into effect but this will all be Backward Compatible so existing LC all interfaces will the same we are doing changes to the LC interface as well so would welcome you guys to kind of come in and give us more feedback in terms of the attestation Bridge we are trying to right now it's an optimistic design but we are trying to make it a ZK not best uh the initial prototype is ready we have already deployed it on Goldie we are trying to you know polish it off along the fing and um we will make sure that it works even with a very large Val set like housing value yeah that's all from me but if you guys have questions feel free to visit our website the dogs the blog um and yeah yeah also hiding thanks