Centralize Block Production for the Execution Layer | Namik Muduroglu - MegaETH

hey everyone thanks for joining us welcome back to build Squad stage here at eth Denver next up we have namk he's the chief strategy off officer at Mega eth he's going to be talking about centralized block production for the execution later welcome to the stage thank you so much yo what's up guys uh packed house as usual um who here has heard about Mega E put your hands up oh [ __ ] yes okay this is going to be fun I I mean people have like all kinds of pre-existing assumptions uh but we can walk through a few of them so off the bat like what's the thesis right which is no blockchain that's homogeneous IE monolithic IE the node does all of the tasks associated with the blockchain will ever scale proportional to world computes right and what we see is just tradeoff Matrix e super decentralized very lightweight amazing no one wants to use the L1 salana is you know a bit less decentralized a lot more performant and what you're seeing is a race to zero a great example is like Fogo which is an even more centralized layer one and that's cool they're really performant and you start asking yourselves questions around why are l1s fundamentally have these constraints and the answer is simple it's consensus consensus fundamentally leads to Performance uh degradation the more decentralized you are right the more nodes participating consensus the longer physics kicks in and the less performant you can be and that's not good uh I think we have the which is smart contract systems have never actually found true pmf that's why Bitcoin is doing a much better than everyone else and what we're seeing actually is that Salon is not very cheap either I made this a very long time ago this deck but these days salana costs a lot almost as much as eal one now that's because there's like 2,000 real TPS on soul and like on E but tldr applications have not ever been able to really sit in the same block space comfortably meie solves this we've built a very cool system where we effectively build um specialized nodes each doing specific tasks for the blockchain and the end end result is similar to what vitalic RADS about back in 2021 in his article endgame where you see centralized blog production but maximally decentralized block validation I'm happy to kind of walk through a lot of this with you guys um or would you rather learn about like use cases and applications and all the reasons why this is cool should we do like some hands who wants me to go very deep into the tech okay a couple awesome so let's do it um so basically we have a thesis right which is you have to get rid of consensus you do this by effectively hypers specializing for execution what we do is we are like traditional blockchains are scenarios where every single node does all of the tasks right and this ends up creating a pretty annoying let's see if I have this on this deck yeah okay I don't so what it ends up doing is you have this weird situation where if each node has the same hardware and software configurations and they do all the tasks Associated for blockchain you need to make tradeoffs right if you want to have a blockchain that's really really good for decentralization I.E validation that means it's going to be super lightweight right well that sucks for sequencing and execution you end up in a scenario where like trade-offs are always made and it's kind of the name of the game for us new data enter the market in the form of fraud proofs and force inclusions Force exits and that basically told us hey no matter what if the sequencer is censoring you you can force exit back down to L1 and we saw this a few weeks ago or maybe a month ago with the sonum L2 when it started censoring user transactions that were you know breaking their IP rules and it were able to go back down to L1 so we take this piece of date and we're like cool what is the most perform first architecture we can build and then we come down to the conclusion that we need to specialize each task for a particular node so what we do is we build effectively a particular node that does all sequencing and execution right imagine an unbelievably beefy node way way way more centralized and way beefier than salana one sequencer we never decentralize it we never introduce consensus and we use all kinds of cool optimizations and techniques so think like inmemory Computing jid compiling paralyzation the whole shebang now we have a really really performing piece of the puzzle and because our whole feis here is that we're going to build specific nodes for each task we now need to build new node architectures for everything else right so one example ends up being like how do we approach validation what does our approver Network look like if you have a super beefy sequencer that let's say costs $11,000 to run does it now cost $11,000 to run the approver cuz if so that doesn't sound too decentralized to me what we do is we use a technial status validation we have a subset of the provs that this Prov read a subset of the sequence's transactions tldr we've optimized the entire execution engine for what we know to do really well and we Outsource what we don't do so well so da is a great example right the biggest bottleneck of existing l2s is there's a pretty strong da bottleneck right so we use aan da um and obviously for consensus settlement we don't want to do that that's really difficult so we just fully Outsource that to eal1 with the new guarantees we have right which is effectively the power of fraud proofs and force exits uh yeah so I kind of spoke a bit about this but traditional blockchains one node does all tasks meie is like yo let's break that up into many different nodes um yeah so what we see here is effectively the node and Hardware tradeoff landscape it's kind of what I described before provs are very very lightweight right and you want them to be lightweight why because you want them to be decentralized meaning anyone can spin one up now sequencers are going to be really beefy and that's fine because for you to build a truly performant environment and performant engine you need pretty high Hardware requirements and you see the entire range uh in this little visual why do we make these design choices is kind of what we discussed before which is proofs and slashing is really important to effectively enable the sequencer to be following these rules Mega Eve will be uh a test net and Main net will be optimistic rollup day one but we're working very closely of all of our very smart friends in ZK land you see we like to Outsource stuff um on basically doing uh proof validation with like a ZK system we think the guaranteed rights of force exits are very powerful freedom of choice and ability to rotate to sequencer so naturally you're probably sitting here saying okay cool these guys have basically fully centralized block production meaning like namk is going to be running this beefy sequencer and sure my transactions can be forced down to L1 if he decides to censor me but like what does this this doesn't sound like a really good long-term solution for a couple of reasons right so we are going to be hitting 100,000 transactions per second on Main net that's 10 gigas and 1 millisecond block times that's really really great for collocation high frequency Traders all all of these individuals how does this actually impact you when talking about endtoend latency so in the land of endtoend latency one of the biggest issues is going to be like your geographical location right where you are compared to where the sequencer is right now the meie sequence here on test is in t in Texas so we can probably enjoy you know 50 Ms N1 latency that's equivalent to you playing Minecraft and having seamless UI ux now naturally if you're sitting in Tokyo that's not very fair you kind of have a horrible outcome so what ends up happening is we're going to be rotating the sequencer to follow economic activity uh across the world right so the sequencer will be following the sun this is a novel technique which allows us to effectively capture major pools of economic activity right it's it's quite simple if it's 11:00 p.m. in New York I guarantee you maybe a couple of us are buying shitcoins but the vast majority of people are asleep while a lot of people are awake in Tokyo and they're doing relatively important stuff right so this is actually a system that mimics FX markets and we effectively will be having three main geographic locations Asia Europe and the us and we'll have a set of rotating sequencers between these different locations yeah and why is this good because now namk is no longer you know uh running the solo sequencer there's never consensus in the system at any one time only one sequencer but it's able to rotate both for better uiux better Capital efficiency and uh removing keyman risk uh I guess here's another visual of the system design um you kind of are able to tell to what extent we have kind of rebuilt some stuff but more importantly which nodes do which tasks right and here you can really see the idea of node specialization to its core we really think it makes sense given the new design choices associated with like more heterogeneous systems for you to be able to off Source uh a lot of the tasks associated with like nodes traditionally so yeah I guess how do we do this uh a couple things I first we're evm equivalent which is pretty sick because a lot of people aren't we rewrote a new state try we bit built out our own database effectively just in time compilation efficient State sync streaming evm this is really cool because it allows us to actually enable real time applications and obviously parallel parallel execution once we have really good single Fred performance perance um yeah so this sounds all fun and all but like you've fully centralized blog production that's kind of concerning uh is there going to be any unlock here I hope so uh our thesis is that by effectively making these design choices we're going to be able to enable a level of performance that we haven't really seen before in crypto we'll be a ble to get web to py with crypto economics with permissionless systems and actually accelerate adoption of just genuinely novel applications so let me show you guys some simulations from my co-founder um what what's the situation so how how many blocks is what's the block uh block time on salana these days 450 Ms can one confirm yes or no no one knows Pablo do you remember okay uh let's just do let's just do 1 second for now because a lot of people are like aiming for 1 second block times right something like that if it's an L1 there's no network propagation that just means you know communication to and for from the sequencer this is what it looks like when you I do something on my laptop on the blockchain of one second second block times and how that's going to be represented on train the bunny is supposed to follow the rabbit right it's pretty [ __ ] slow that's that's like good by the way that's the goal one second is the goal in for a lot of teams right so I believe uh base just announced something pretty cool called Fast blocks did anyone see that today there we go bullish I think that was 250 Ms okay close enough and because base is a single sequencer you have to deal with network propagation so let's pick a random number and uh we'll keep that for like all networks that have no consensus okay guys so let's just do let's be generous and say 50 MS right let's pretend the sequencer is close to us okay sure but imagine if you were playing Minecraft you'd be pretty pissed imagine you were playing a MMO imagine you playing Call of Duty like you would rage quit right okay so now let's look at Mega for context the meie test will go live next week we are sitting at 10 millisecond block times for the vast majority of transactions let's just do 15 cuz this is complicated and we're sitting at 1.3 gigas oh wow that feels pretty smooth in my opinion but I am a humble I don't know much right small brain so I think this is pretty cool and our whole feis has been like cool we've made some crazy design choices but we better find insane apps that will enable like we we'll find this unlock right so we started a program called by the way this is our Echo deck so I'm a bit lazy so this is not fully relevant but we created a program in which we had a bunch of Founders that are super cracked we made big bets on them to build applications which are only possible in meie they all leveraged that unique uh technological unlock of 10 millisecond blog times to build real time apps these numbers have changed in the past few months we now have 17 ecosystem teams who have raised around 35 million right and until our recent nft sale the applications on meie had raised more than the train than ourselves which I think is pretty cool e did that but I don't know anyone else right that's I don't know pretty cool um okay cool that's it on my end uh I'll give you guys back a few minutes but huh I I don't know why I just disconnected the HDMI cable so but I'm pretty excited we'll have test that next week and it's really cool there'll be some interesting applications if you're a builder and you want to test a chain please come and see me because I will give you everything you need to chest a chain and it's exciting the benchmarks are good and I think like we've made in my opinion the most logical construction for a sement layer like etherium ethereum maximally focused on being the most decentralized turn complete blockchain in the world in my opinion existing l2s chose to replicate ethereum's design choices and design principles we didn't do that we said okay let's take this new data and then's Sol for what we is weak on which is performance and we built the most logical execution engine by decoupling consensus from performance and we're excited I think it makes a lot of sense and yeah thank you so much for your time today guys