L2 EVM Common Core: A Path Beyond EVM Equivalence by Ansgar Dietrichs | Devcon SEA

[Music] hey everyone um yeah I'm here to talk about the L2 evm Common Core um and it's basically this um initiative that's a recent initiative that came out of the Roll cor process and so if you've never heard of the rore process that's uh something we started earlyer in the year and it's basically a al2 um coordination effort um we have a monthly call um where um many of the of the major l2s on ethereum sent representatives and uh it's basically for now mostly used as like both coordination across them also um a bit of a trying to be a bit of a connective tissue to the L1 I'm personally most spending most of my time in L1 core development um and then we also have what we call rips rollup Improvement proposals and the idea there is basically that they are um similar to eips eips you might be familiar that they are um proposals to add new features to the evm um on L1 and so that the IPS are the equivalent for for the L2 side of things um and yeah the process was originally kind of started the roll core by um Carl who can't be here today unfortunately y off and um and me and then uh more recently Nik also uh started um joining us more actively um and yeah I want to basically talk a little bit about uh evm equivalence in general and then kind of the specific kind of direction that we see the evm equivalence go uh in the future and I want to start a little bit with history um it it's going to be a bit more of like a easy kind of beginner friendly talk at least in the beginning um I like pictures so I have some pictures um in principle the way I always think about scalability for blockchains um um is basically as this fundamental trade-off between the the C cost to run a full node and the throughput you get right and so Bitcoin ethereum to early examples ethereum has more throughput than Bitcoin but it's also a bit harder to run a an ethereum full node than it is to run a Bitcoin full node right and I call these kind of trustless chains because the important thing is everyone on Earth who wants to can verify the chain at least that is that is the goal and uh we are sticking at Le pretty close to that um and then there is the kind of the um approach of just going further up the diagonal and just picking picking um uh Solana as an example here uh of chains that basically just do just say okay we want higher throughput and how do we get there well we basically go up and we accept that now it is harder to run a full node but we get more throughput and of course the trade-off is that now um you actually need to have a quite beefy machine to be able to still trustless participate otherwise you have to trust the majority of of nodes um now ideally we want to go there right the ideal chains like they have very low cost to run full not everyone can verify but they have high throughput so how do we do this um we basically this is the the key rollup trick right we realize that actually what matters is the cost to verify the chain and if there's a way to verify the chain cheaper than running a full node that's sufficient um and that's basically how we how we get to Roll-Ups because rollups are basically a way to compress verification of a chain and so what you can do now is once you get to that place you can do either you know the the basically the thing you could you could immediately think of is just hey take the high performance chain and just run it as a rollup and there are um some projects now on ethereum that that do this that are Solana VM based um it's it's a bit of a challenge though to adapt them for el2 then of course what you can do is as well you can just take the evm and you can really scale it to high throughput and people sometimes have this mis mistaken conception that like the evm is fundamentally not able to go to similar throughput levels as the svm that's really not the case the reason why ethereum runs on Lower throughput is specifically because we want this tradeoff we want to be verifiable by everyone on Earth in a trustless way right it's a choice it's not a technical constraint so once once you go into the rollup you can actually push the evm to its limits but the um variant of this that we've seen the most is actually what I would I call I don't know the multi rollup cluster which is just like many smaller like lower throughput individual rollups that basically together form like a very high throughput system um so that's kind of like the background of how how the the uh ecosystem evolved like as ethereum went uh through its Ro centc growth map and if you zoom into like now the chain itself and then eventually the evm I have some more pictures I call basically this that this just like the execution chain that's just a symbol that's representing chains where actual activity happens right like chains that actually offer execution where you can build apps on top and and the most basic version of a chain is just like a self-contained chain it's like an L1 with just normal deps on top and but now you want to turn it into a rollup how do you turn it into a rollup well you need settlement and data a ability um and those two things can be provided in a combined Way by a settlement chains um so then if once you have a settlement chain you can just take take your your chain there and just turn it into a roll up and the nice thing there is that now you inherit the security of the of the settlement chain but you can still now yourself run with much higher throughput right but because you have that um that security that you inherit from that low throughput but high security chain that's kind of like how how this kind of this best of both World solution Works um and that's kind of how you get to this world right where you have a settlement chain and you have a bunch of rollups on top now this was basically the original vision for what people um back then called e two you might remember before the merge people used to call kind of like the beacon chain eth 2 and we moved away from that name and that has a specific reason why because um E2 was this Vision that we would basically uh give this new chain some fancy new execution and settlement system and we would take the old eth1 chain and the old evm on top and we would retired back in the day there were a bunch of different thought experiments you could turn it basically into some sort of rollup on top of the new E2 chain you could just have it parked somewhere on the side with some you know either like proof of work forever with some bridges to the new system or just even over time um uh deprecated or something but then at some point basically as we moved closer to the and to to these ideas people realize that hey actually the eth1 chain is kind of good enough for what we need for the settlement chain and so that's why then instead the plan became let's merge the existing system into this new E2 chain and that's why we kind of stopped calling it E1 and E2 and instead just call it ethereum again um and so that's why today what we have on ethereum is this hybrid settlement and execution chain right so like down there it's that's what ethereum is supposed to be like that's both the settlement things settlement and data availability but we also still have execution you can you know have your EA on on L1 you can have you can use Unis swap on L1 all of that still works on L1 and it's all still powered by the original evm that we used to have on ethereum from the early days um right so that's basically like how the role of the evm evolved and for how basically how for a long time the evm was basically supposed to be to have like an expiry date and then it turns out it it actually didn't um now how do l2s and their execution systems fit into this well actually when we first pivoted from sharding to roll ups the idea was well this is going to be amazing because now all the l2s can go on this wild exploration and see what type of new fancy execution system works for them works best for them but then it turns out we have seen a little bit of that and I said that earlier we now see experimentation with the svm and move VM and like more fancy High throughput EVMS and whatnot that happens a little bit but the vast majority of l2s actually basically followed the network effects and they were like well there's already a lot of dep Developers for the existing evm today so what we're going to do is we're going to just copy the evm right we're going to copy the evm and we're specifically not going to make changes to it because otherwise apps would not be compatible um it doesn't mean that the l2s were not Innovative I would say quite the opposite it I think we've had we've seen like 3 four years of pretty rapid innovation in the space but that was very much focused on the specific specific new challenges for L2 so that was focused on um um like new functionality and and uh like you know like fault proof ZK proofs of course on the in the ZK space the innovation has been has been amazing to see um and then when it touched the evm it was more like extension so I'm not sure how many of you are familiar with arbitrum stylus that's an example of an extension to the evm that left the core of the evm unchanged and just added something on top um and then one more wrinkle to this the the consequence of the evm equivalence was also that L most l2s ended up just going with the dominant L1 evm client which at the time was G of course G is still the most popular one by now there are also others that are a bit more on equal footing but back in the day really G was the only main one and so it for for for that reason is still the the very dominant client in the in the L2 space so that brings us to the present and this also brings us to the point where like the pictures part of of the presentation it but because I got so much feedback in the past that my presentations don't have enough pictures I decided to just add one extra picture to every slide from now on so you'll get a few bonus pictures starting with this cat but um let's focus mostly on the content so today where are we at we have the evm on L1 and and basically if you look at uh the kind of new features that we've shipped on the L1 we have roughly one hard Falk a year you can you will see that we don't really ship new features to the evm that often right like it's it's a pretty low frequency we've had a few more what I would call ambitious features um in the past and by the way this is a good point to briefly mention I say evm a whole lot throughout this talk what I mean is technically not just the evm like the core evm itself but like the execution system as a whole which is a little bit more expensive some people are always like oh but state expir is not technically part of the evm Yeah so basically when I say evm think execution in general um but we've had basically a few of these more icious features that we worked on and actually basically got all the way down to specs and the ability to in principle ship them and then rejected them so that was two of them that I worked on in the past were state expiry to manage the state size on ethereum and Native account obstruction and both of those were ready like 3 four years ago and we just decided hey it's just not a priority for L1 um and then even when we ship new features to the L1 uh which we still do admittedly and like um um there some of them are actually quite exciting we still very much focus on what the L1 needs right and so that's specifically for example I 7702 which I'm quite excited about um as one of the co-authors which is giving some smart account functionality to existing user accounts that is uh euf which is upcoming which also I think is is actually a little bit underhyped I mean I guess you you have probably heard the term but um I think it's going to be quite amazing and then a little bit further down the road something like veral trees it's not 100% decided if it's quite going to be veral trees or some variant of it but but basically um that is in the pipeline but it will take several years to to arrive um and it's very much uh something that that primarily is is is interesting and and nice for the L1 so so how where does that leave the the L2 well it turns out that most of the l2s that started out as equivalent to L1 basically now are stuck in that position right the the point is that you don't want to be the only chain that starts deviating starts implementing features that no one else has because then none of the wallets none of the tooling none of the death tools um are compatible with you right and so what that means is like now these l2s in a way are stuck with the evm of L1 and are stuck with the kind of the slow speed of these new features that are being shipped um and in particular it's not just clearly the L1 is very conservative ships features in general not not quite as fast as l2s but it's also a fundamental difference in um constraints and priorities so uh it's about throughput Ambitions the difference in full note requirements that comes back hopefully to the graph right if you remember the how basically the rollups give us this Best of Both Worlds you can run these very high very demanding very high throughput chains and still inherit the security of the L1 well but that also means that you're building a system for very different requirements um block building works very different on on L1 versus L2 and then of course for the ZK rollups ZK basically changes everything in terms of how expensive different operations are and uh how the VM should ideally work so basically all of this means that like the evm is just not an ideal fit what for what the l2s need um also G is still the main client uh on the L2 side but G is very much focused on the L1 like the team very much has made it clear over the last few years that um while they are you know they're very happy that that they get so much adoption on the L2 that's that they very have have very limited bandwidth to basically try to provide L2 specific features but that's the Silver Lining here we have seen several l2's several other clients gain more adoption in the L2 space that are more willing and more focused uh on that use case natively so how do we where do we go from here how can we basically make sure that we give l2s the best way to to really make the most of this kind of this better this this Best of Both Worlds trade-off space there where they are in this unicorn range of of basically uh high throughput with with low trust requirements like how can how can they Bas have have an optimal system designed around this well SE two pillars basically two pillars that that we have been focusing on first one is just make sure that all l2s have a client that actually is willing to focus and support the L2 use case more natively and so that specifically tried to identified and tried to eliminate um this bottleneck of of the of the L1 focused guest team not eliminating the guest team just eliminating that bottleneck um and out of that came the collaboration with uh nethermind which is the what we call the rollup gu team or what they call the rollup gu team which is a fog of gu it starts out identical but over time we'll add more and more L2 native and L2 specific features and will basically join the set of L2 focused clients right like nethermind also you they have their own client so nethermind now has the nethermind client and the rollup gu client both of them happy to support l2s then there's also WTH there's baso there's Aragon so there's we have now a basically an enti entire team of clients that are willing to to basically support um l2s um and basically ideally that way we can um we can basically Leverage The the existing kind of uh skill set that that that we already had in the in the nethermind team and and and and really kind of help help el2 there so that's pillar one pillar one is make sure that all l2s have a client that that is willing to support um rollup specific features now um that yeah as I was saying that that is basically um joining the kind of the broader set of of of clients and now we can basically start start to ship more IPS more ambitious IPS because the clients can support them the problem is that on its own that's why it's only one of two pillars because that on its own does not get rid of this kind of fundamental lock in of the evm equivalence the equivalence to the L1 constraints and so that's where this second idea comes comes in that admittedly is a bit more opinionated this one might or might not play out that's something that we've proposed to rollups just over the last month or so and there's a lot of interest in this idea but it's not clear that that will actually play out so that's basically the proposal that I wanted to to talk about here today uh the idea is I call it the uh we call it the L2 evm Common Core and so the idea here is to just say hey basically let's just start like what basically what you really want as l2s is you want to be equivalent to each other right because that is where the network effect is and you don't really care so much that what you are equivalent to is exactly the L1 so what we want to do is we want to take the L1 evm and over time extend it more and more and have that be the share Target that everyone focuses and everyone supports um and that way tooling solidity all wallets all these kind of uh this the full stack can now start supporting this L2 Common Core as as the base uh feature set and we can really start shipping more ambitious features uh over there and um yeah just just to to illustrate that's I count that as an image um that's basically like how you can think of the feature set of a rollup right so you have the L1 evm as the core everyone supports all the features of the L1 evm in the future there would be this extra ring which is the L2 evm Common Core that also everyone on the rollup side would support then of course you have stack specific features so you if you are on optim one of the op stack chains then that would be op stack features if you're on polygon that's the polygon features Etc and then every chain also has some features that make it unique right and so that's kind of the outermost ring and so what kind of features are we talking here like what could we bring well basically the idea is really just let's open the design space again right now that we have a platform to basically ship more ambitious features let's just look into what what what we have there so in the short term that could just be things like basic repricing that to make it for example more ZK native a 2d transaction type that something that I uh think would be really nice right now the way l2s charge users for the settlement cost is differs from L2 to L2 which doesn't make for great ux I've sent a transaction on optimism once and my wallet said at most it's going to be 1 cent I sent the transaction and it's 10 cents and I mean yeah it's still only 10 cents but imagine if that was would be fact to 100 you know like all of a sudden it would be not so funny that that basically that that doesn't work so having a native transaction type for l2s would actually be quite fun um delayed State Ro computation is maybe a bit nerd it's it's a feature that that people have been talking about a little bit that could basically make uh blog processing more streamlined um and then in the long term that's what I'm really excited about these kind of features so that would be things like native account abstraction I'm not sure how many of you all have a smart account have heard of ERC 4337 in the past and there are proposals to basically have that be natively embedded into the protocol eventually I think this will also come to L1 but it will take years and years we can ship it in the meantime on L2 transaction priorization state expiries statelessness State rent like basically figure out a way for how we can basically um manage State and and and and state growth more more aggressively and then something I personally am very excited about because I've been working on that in the past is multi-dimensional pricing um I think we're basically out of time let me see I think I only have like two quick slides so let me just rush through them uh next steps again the common core is a bit more unclear yet like that's a proposal we will have to see how much interest is there really and how much willingness to commit to this across the ls figure out a governance because given that that would be a shared set of features that everyone would commit to supporting that requires governance among the participants um and then have a have a research process across across the teams um and that also would mean that for roll call in the season two of roll call we would basically Focus also on facilitating that side of things more so that's all I hope that gave you some inition around why that is at least an interesting Direction and why there is some uh use in trying to eliminate this bottleneck of the equivalence to the L1 which really has very different constraints from the l2s thank you okay cool let's Scurry towards the middle because you'll have we have some questions that need to be answered so starting from the top can you quickly explain why eof is underrated yeah um that might be a question from an UF hater I can I can already feel that which is fine you know I think UF is one of those features um where not everyone loves it I person name the most excited about is specifically in the L2 context actually and I guess that goes a little bit counter to the slide that said it's a L1 a specific feature but because the nice thing is it gives you version bite code so so far on ethereum we only have one type of contract code so if you deploy something it will always behave the same way no matter what with UF contracts can now have a specific version which means you can basically have different Behavior depending on a version and so that means you can have like backwards incompatible breaking changes new types of new features um uh we have deno in the audience Deno is like the expert on UF so if if anyone has like actual interest good faith questions about UF go talk to him afterwards he's like the Bible on on on UF but I'm I'm a fan okay what is the best out VM from your view o SP I see I mean I don't really have an opinion I think I'm very very excited that we now have this Cambrian explosion of l2s I know that some people on Twitter are like oh you know yet another L2 yet another L2 I actually think there will be like we'll look back at this time and we'll be like oh we only had 100 l2s you know now we have 10,000 so I think I think we we'll get a lot more l2s because before we get fewer and I think that's a good thing I have no idea some of them will work really well most won't um and then we'll see which winners merge um I mean of course I think I do think of course one interesting direction is svm based ones just because you know Solana is a very interesting kind of alternative Network I think as an L1 it doesn't make so much sense I want to actually be able to verify that someone is not stealing my money but as an L2 software St I think it's an interesting direction to explore do l2s ALS oh it's changing up votes are happening um but I'll start I guess two votes win will l2s remain able to execute contracts written for L1 or will compatibility be lost yeah that's well that's great because that question shows me that the person actually was thinking during the presentation about like the educ like the the the implications of of of this idea like that's actually exactly like a really interesting question like will we be able like how close will L1 and L2 come in core if if it if it starts to diverge how close will they remain H I would predict that for the next forceable while probably you can write the same code for L1 and L2 I think at at some point we might reach the point where L2 might want to ship features that unlock new capabilities like parallelization like as like like asynchronous communication within the same block or something that might just make it impossible to keep the um uh the compatibility but we will see if it's if it's worth to break my my my my prediction would be long-term longterm L1 and L2 will probably diverge to the point where you have to write L1 specific code if you to deploy on L1 interesting do l2s also oh my God things keep changing um will nethermind make rollup nethermind well the fun thing is that rollup nethermind is just nethermind like the nethermind client is very willing to actually focus on uh shipping features that are useful for L2 specifically so the only client that needed a rollup upgrade was G specifically okay we'll finally get to this one do l2s also aim to have client diversity yes great question and that is also even though say for example more and more rollups are starting to adop breath and nethermind and these other clients and you could just say well why even bother with rollup gu because you could just wait until they all switch over to one of those clients that is more native natively um aimed at the challenges and the the kind of the the features that the L2 want but the answer is actually it's very very valuable even if they start supporting these other clients to keep um G compatibility so that you can go to this multi client world because say for example I mean already for fault proof it's very nice but then also with CK proofs it's very nice if you have multiple kind of independent paths that you can prove something from so then you can have like a two out of three ZK kind of disagreement rule or something um all of that is unlocked by multi client uh and I think L2 is on a path towards multi client okay perfect that right