Standardizing How Apps Read and Compute Data Across Chains | Bryan Pellegrino - LayerZero

all right hey hey everybody uh going to be talking through two things today one being uh talking a little bit about layer zero read uh why we built it uh what it does and how it's used currently in the ecosystem and then sneaking in qdb uh because that talk was vetoed later on so we'll just squeeze them both in so layer zero uh who are we right now what do we do um largest uh largest interoperability protocol currently in the space 120 plus blockchains 144 million Messages 75ish Plus billion dollars of value transferred closing in100 billion which is awesome uh 71.3% market share when it comes to just pure messaging volume 60% market share when it comes to Value transferred and the thing that's interesting there is that uh um the market share Delta between those two means there's a not at all insignificant percentage of the messages that are starting to be nonv value transfer um which is which is really interesting and so what is layer zero read read is the ability to take any data from any chain over any period of time and so you had this problem historically if you wanted to access any state data from say ethereum uh originally was just you you couldn't get it you needed some external Oracle uh to provide you that that data uh most people can't get these oracles to even respond to them let alone on like a very long period of time so this was like hard uh to access and then it was even if you could access the data it was both slow and expensive so you can imagine trying to fetch um something from ethereum who owns a pudgy penguin or a Bor AP uh what is the current price of the eth usdc pair on ethereum um on Unis swap V3 any any of these pieces of data you would have some latency coming out of the chain you were requesting it from uh you'd grab a moment in time snapshot of what is the thing at that exact moment which is obviously kind of manipulatable to price to you know uh attacks on uh on price or just like Moment In Time snapshots and then you have this 3 minute and 15 second latency coming back to deliver the data and so not only uh did you have pure Moment In Time data you had really stale data but the time that you needed it um so read is designed to sort of address all of that any data any chain any period of time so what are some of the use cases uh one of these is election Market uh that got settled in the last election winter MW introduced um basically the ability to create any Market in a prediction Market style uh but they need a way to like resolve what actually happened there they used Edge chaos uh Labs as Edge Oracle for this and so the election happened there were markets that lived on any of these chains so any anybody could buy either side of this Market on any of these chains uh through just like an amm structure uh Edge uh through chaos post the election result onto arbitrum and then every other chain just uh absorbs that data basically through re to settle all the election markets simultaneously um Shadow assets is something that we see more and more now for anything that needs off or ownership so you can imagine a world where uh your authentication lives on a given chain this could be World chain uh again this could be the ownership of who owns this this board ape or this penguin one thing we saw that was cool uh was ape chain itself they didn't want to actually transfer the ape right you don't want to be bridging this asset because there's a bunch of stuff that was built on ethereum that relies on those ownerships you have some um some traits or some benefits that you've gained in that ecosystem and to need to move the asset uh might actually like harm the end user so this Shadow asset was the ability to prove with Reed that you own the asset without actually being able to move this and we actually saw this in a couple of games uh prior uh some some of the biggest games um proof of play pirate Nation a couple of these uh basically what they really wanted is hey we want our uh assets and our ecosystem to live on on this chain we want people to play the game we want people to have um everything that they own here but all the liquidities on ethereum right so like we want the real assets to be traded to live back in ethereum and so what they really want is kind of like an ad hoc Oracle that's a representation of of what what what do you own and then when that changes you can just there's a couple of really interesting ways you you I won't get too deep into it you could use a bloomfilter do some really cool stuff here but um basically without needing to move the asset back all the time without needing to touch ethereum State and obviously touching ethereum state is the expensive part back when e transactions were $15 to $20 again you want to fetch a single piece of data like hey who owns this nft or like somebody just sold this you're going back you need to do a right on ethereum to request what you're trying to fetch and then another write on ethereum to send the response back so you're paying you know $40 to $60 in ethereum L1 fees just to get the response of what you're doing with read you don't ever actually touch e State you're never writing anything it's read only uh and so you write on your extremely cheap chain that costs one penny hey fetch me this data or this construction of data and then you immediately get the response back next block right um so there's no latency there's no cost you're not touching ethereum State at all and so what that enables you to do um there a bunch of really cool things but one you can do a lot of stuff that you you know would cost $60 before so the the world of actions you can take when you're paying 10 to 60 bucks per action versus the world of actions you can take when you're costing you know a penny per action is just night and day difference and that in its own uh is super exciting so one of the things I was most excited about was this ability to stream pricing data so you're going imagine um not just being able to grab what is the current price of eth usdc on uni V3 on ethereum uh which is great but I can grab a 100 block T and I can grab it of any longtail asset on any chain and I can live on uh some new chain that's spun up right we have 120 plus chains right now uh I promise you the you know 118th of them uh is going to have a really hard time uh getting chain link or getting a Oracle provider to give them price feeds consistently it's just always been a problem within the industry this is the ability to be able to build constructs around pricing across any formulation that you want and you can grab yeah what is this um you know what is the price in this uh decks uh on salana but you can also grab like I want a 100 block t-w across these four exchange these four dexes that live on four different chains and then I want to like amalgamate them all and and create some price index based on that and anybody can do this permissionless there's no uh you know nothing custom that you need to write you don't need to ask us to add a asset you don't need to ask us to do anything you can read any piece of State on any chain in real time and you can apply the sort of external uh mutations to them which is like really really interesting in terms of what you can build and what that enables in like long-term defi um so yeah that's Reed uh super excited about Reed have seen usage uh slowly but steadily ramping up we did it for our own voting I think you'll see more and more governance across multiple chains use this structure uh more people like AP chain like a a ton of others are starting to build their own internal Primitives on it which has been cool to see um yeah I've spoken about for a very long time I've always been much more excited value transfer is interesting but honestly when we built layer zero there already Bridges right Bridges existed the entire point of layer zero was that everybody was thinking about these things as purely a bridge and for us the Insight that we had was value transfer is just a special case of generic messaging right uh really what you're doing is just sending on the Internet it's arbitrary contract invocation with a btes array and the ability to take that and express it in any form that you want uh should unlock a ton of really exciting things especially as applications become more asynchronous more um data Rich across multiple environments and for the last couple years it's pretty much been wrong right like almost everything has been value transfer uh but I think that is is quickly changing and I think you're seeing applications become more uh stateful at the application layer I think you're seeing applications uh start to lean into not becoming just a single isolated instance across a lot of different chains uh but becoming like real asynchronous instances Stargate was like one of the first examples of this um yeah so I'll dive into qdb now which is the the talk that uh was not supposed to be a talk so uh this I think is is for me one of the more interesting things we've worked on lately lately and like totally orthogonal to Pure uh pure layer zero right um we've spent a lot of time probably more time than any other team in the world in actually digging in to every single VM uh that we have to go to and so like we built ethereum and we've always taken the stance that the protocol must be immutable the protocol is not immutable then every application built on top of us has to rely on us for some layer of trust and that is not an acceptable precondition and we don't think we ever would have grown to the size that we are now if we didn't have that uh but what that means is every single thing that you do is terrifying because you get to basically launch the protocol once you cannot fix it if there's a bug uh you know this is the same way that Unis swap has built everything for a very long time so it's really a measure twice cut once and so um you know I I talk regularly we spend probably $5 million a year on audits I think we're the largest consumer of audits uh out of out of anybody uh our ethereum contracts have been live and in production you know since March 2022 uh so going on three years now those are battle tested tens of billions of dollars on them tens of billions of dollars flowing across them launching to a new evm is quite trivial right it's largely of fork of the same code launching to a non-m is completely non-trivial you talk about our salana program compiles to 50,000 lines of rust uh which I think is one of the largest salana programs out there we talk about ton we just went to um you know we literally rewrote an entire programming language fun C++ on top of the ton VM uh in order to build this and help other people build more easily to to Aptos move VM like many of these VMS we've found crits uh We've helped teams patch them uh we have to go really really really deep because it it matters on the security side in every single layer that we go to and so in doing that one of the things we've seen is kind of what um stresses scaling uh in each system and if you look at Aptos his recent chardin million TPS uh example that they give storage was the bottleneck there for each of The Shard right each one of the chardin had 33k TPS but we're bottlenecked by storage so uh what what is the Insight here uh and a lot of this Insight actually came from our our Chief Architect was on paternity leave and he just started poking around and uh I guess cuz this is the thing that you do when you have a baby uh you create a really interesting dat Bas so conventional belief was you must use a searchable data structure I.E a try and everybody built that way and the challenge is tries uh have really significant right amplification uh which caused a lot of issues and the alternative in the Insight that we had is that actually a log based DB that's modified to support mutability exclusion proofs and historical proofs are is actually extremely highly conducive to inmemory merization and so what does that actually look like right uh it's fast really fast 3 million updates per second uh and 3 million updates per second is uh well I'll just look compar to state-of-the-arts right now even for unverifiable databases databases you are not doing merization at the DB level this is 6X faster than an unverifiable database while doing merization if you look at state-of-the-art for actual verifiable DBS which is nomt uh it's 50x state-ofthe-art and it's 92% theoretical optimal on current evm workloads for dis iio uh which is just like an instain step function there is no 10x more from here in terms of what can actually be gained it's it's it's really incremental and so uh this is a huge step function we've seen a way more interest than I expected in terms of teams uh wanting to adopt this because storage has been you know you think about any architecture uh computer architecture block Chan architecture it's really storage compute in in network right uh and this really uh creates a lot of compression and solves one of those uh one of those vectors um and so if you look at overall highlights 2.28 million stayed up Ates per second over a million TPS uh benchmarked at transfers per second in terms of token transfers uh Benchmark with workloads up to 15 billion which is 10x ethereum's 2024 State and have proven capacity to scale to 280 billion entries on a single server uh so really massive massive uh gain there in terms of what can actually be held and processed um single read per state access so order one uh reads which is anybody who spent time in computer science uh it doesn't get any better than that an in-memory merization on a footprint as small as 2.3 bytes per entry uh efficiently scales across both consumer grade and Enterprise Hardware so small machine is going to be as efficient as a big machine it's really a lever that you can play with uh and available for any blockchain to implement uh everything is fully open source uh Apache MIT uh completely available to the world uh received a lot of amazing uh feedback from within the community from Uma talking about what this actually means for for real-time blockchains and realtime proving uh relative to needing to pack an entire block with MPT and having slow proving on top of this so this uh gap of latency that's being shifted to being able to pack on a per transaction basis with qdb uh Patrick I'm a huge fan of at common wear uh talking about him you know not believing that this was uh was possible in Yong from Mega eth talking about how we basically you know perfectly solve the bottleneck of merization with this so uh super excited about qdb super excited to share more kind of with the world and see the first group start to implement it and uh yeah more or less hopefully going to keep building uh things that are genuinely useful to the industry as a whole open source as as much as humanly possible and help the space move forward so thank you all so much