 Yeah, my name is Ansgar, I'm going to talk a bit about scaling, but I want to start first just with a very quick apology. In case you were here to hear about multi-dimensional resource pricing, I had a bit of a hard time to make up my mind which of the two talks I wanted to give, so for a while on the website both titles were on there, so if you're really into multi-dimensional resource pricing, which you should be, I can only recommend, I gave a talk earlier this year about it in the Ethereum context, which is quite interesting I think, and then also with EIP 4844, we are about to introduce a, there's an open PR to update the fee market there, and hopefully fingers crossed this will be the very first time that we start moving to multi-dimensional resource pricing in practice on Ethereum, like in this case two-dimensional, so do have a look, but today here I'll talk about scaling and specifically roll-ups, why roll-ups, first I want to talk more and more on the theoretical side, why really is our roll-ups so fundamental to scaling, and then bring it back and talk about Ethereum's future place in that world basically. So first, why roll-ups, and I want to start by with something that I call execution chains, it's basically just blockchains as you know them, so Bitcoin of course was the very first ever special purpose execution chain, special purpose just because it's only for payments and money, Ethereum the very first general purpose execution chain, and in general all these chains of course they want to have maximum functionality, maximum throughput, but that's a tension with needing to guarantee security and decentralization, and to illustrate that kind of the tension I created this final graph, so basically on the x-axis you have the cost to run a full node, and on the y-axis you have the throughput that your network can have, and with Bitcoin and Ethereum basically you are in this nice zone down there where basically everyone who wants to can run their own chain locally, I don't run my own Bitcoin chain, I occasionally use Bitcoin but I could, it's just a trade-off that I choose to make, but that does mean that these chains are fundamentally limited in throughput by basically the minimum consumer hardware capabilities, and kind of the key insight here really is that everyone validating everyone's transaction doesn't scale, but of course we don't want to give up security by kind of moving beyond this, so like the very first thing you could do and I picked Solana just as an example, and this is not meant as a slide at all actually I think in a second you'll kind of see how in a way they're kind of moving in an interesting direction here, but basically what they're doing their version of scaling is just to basically go further up this diagonal and that does give you higher throughput, but of course it means that now running a full node becomes really expensive, so that's one way of just increasing throughput. The problem is, what if you don't run your own full node and there's some sort of disagreement, some malicious attack, some chain fork, you only have two choices, either you go with a majority, in which case 51% of the network can always rewrite all the rules, take away your money, everything right, or the other thing you can of course do is how halt and recover via the social layer, but that's very slow of course. So where do we want to end up on that graph, that's basically what I call the unicorn zone up there, where everyone could trust and validate the network, but we still have really high throughput. So how can we basically find a way to get there and that's basically where roll-ups come in and why do roll-ups come in there, to explain that as you might be aware, you probably are aware of different flavors of roll-ups, optimistic and ZK roll-ups, the easier to explain in this context is optimistic roll-ups, so basically to kind of get an intuition for how optimistic roll-ups solve this kind of this dilemma here, we can look again at the Solana case and as I was talking about right, if there's a disagreement and even if you don't run your own full node, you always have the option to just halt and recover via the social layer and the optimistic roll-up kind of idea here is what if you could just massively speed that up, right, instead of running that kind of this fallback mechanism on human brains, where whenever something goes wrong you have to actually go on Twitter and whatnot, what if we basically just automate that, right, like we replace the human brain with software and in particular with fraud-proofs and that's how you get optimistic roll-ups, so kind of trying to keep with my little picture symbolism, so their optimistic roll-ups are also basically a form of an execution chain, they are their own blockchains basically, the way they work is that they apply the changes optimistically and anyone can submit a fraud-proof, so the only guarantee you need is that there's someone in the system that notices if something goes wrong, but one of the assumptions is a way, way, way less severe restriction than a majority honesty assumption, and then these frauds are automatically resolved on some sort of settlement platform, yeah, that you need, and then the alternative as I was saying, ZK roll-ups, they are just cryptographic magic, right, so in that case you don't have to do this retroactive fixing of problems, instead you just have any state update come with a cryptographic proof that no one did anything wrong and again you need some sort of settlement platform where you can resolve this. One more little caveat here is that basically with roll-ups we're moving away from this world where everyone needs to run their own full node to ensure integrity, right, I could basically only follow the settlement chain and still basically have all the guarantees about the roll-up, but one additional thing to ensure is that we have to also pay attention to data. If you run your own full node for a chain, data is implied, right, because you actually download all the data and execute it locally, but with roll-ups you don't do that anymore, so now we basically need explicit data availability mechanisms, and that's what I put this little kind of database symbol there. Yeah, and with that basically, once you have an explicit settlement system and you have an explicit data availability system, now basically you can break out of this kind of graph of this full node versus throughput basically trade-off, and how can you actually use this now to build full blockchain systems, that's kind of what the second part is about, like how do you actually turn this into something useful, and this is going to be very picture-heavy because I kind of, I don't know, for me it's mostly about giving good intuitions here for like, I don't know, my mental models around this basically, that was my motivation for the talk, so again this is like an execution chain, I drew a little box around it, it's basically in its own box there, it has its own security, and that's the world as it used to be, the blockchain world, we had a lot of these boxes like a couple years ago, right, and now we're adding these exciting new tools to the picture, we're adding settlement and we're adding data availability, and how do they basically change the picture, well first we kind of have to bundle them, or we have to kind of put them into some useful form, the way we like to think about it is bundling them into what we call a settlement chain, which is its own blockchain there, you can see it as its own security, it's not the only way to do this, like if you for example know Celestia, which is a really, really interesting other kind of project that basically works with similar kind of models, they keep data and settlement separate, but for us the way Ethereum kind of likes to approach these things is by combining them into settlement chains, so now we have these two separate chains, how do we basically now actually make use of them, well the way to do this is by, and that's what the errors are for, by basically turning the execution chain into a roll-up, it kind of functions as it did before, but now it uses the settlement chain for data availability and for settlement, so you need some sort of fraud-proofing mechanism, or you need some validity proofs attached, but other than that it still feels like a normal blockchain basically. The simplest way to do this is what we call like an entrant roll-up, so in that case you'd have like one settlement chain and one roll-up on top, and it's basically like a one-on-one to one relationship, and a very simple way you could imagine basically this process playing out in the blockchain world is, and I guess that fits well into this kind of the multi-chain vision would be where every blockchain that wants to basically go hyperscaling just adds their own little settlement chain underneath them, and that would basically bring you to this picture with multiple different blockchains that would all now have their own settlement chain. The downside there is that for one you have fractured security, right? Each settlement chain has its own valid data set, or its own miners, or its own, however you get your security as your chain, and basically some of those will be higher security, some will be lower security, and that also affects bridges. Maybe some of you have seen kind of Vitalix a while ago about how basically if you bridge multiple chains you always end up with the weakest security of all of them, right? Basically if you want to use assets of like 10 different chains, if the security of one of those breaks like that specific asset just basically is broken for you, right? And if you have some sort of defense system where all of this is interconnected basically everything can crash and burn, you really don't want this minimum worst of all security kind of situations. And also I guess that's less of a severe thing, but also in this picture you're kind of duplicating a lot of complexity for each chain. So what would be an alternative vision? How do we think the future might look like? That's what we call like a shared settlement chain. So in this case you have one big settlement chain, and you can have multiple worlds on top. Of course probably realistically the future will look like some sort of hybrid. You'll still have some blockchain that's preferred to do their own settlement chain, but we predict that like this will be a big chunk of it basically, like a shared settlement chain and why is that advantageous? Well two main points really. One is the shared security. I like to think about it like shared pool security. It's not quite free security. Sometimes you hear people talk about how if you turn your chain into a roll-up you just get the security of the base chain for free. That's not quite right I think because actually if you have ten roll-ups on top of a settlement chain then they basically all put additional strain on top of that settlement chain, like now the incentive to attack the settlement chain is 10x as high. So you don't just get like security isn't just created for free, but it is pooled. So now basically you have to break the entire combined system instead of just being able to attack the weakest link. And what that also gives you is then because you have like one shared trust zone basically that all the bridges between those roll-ups if implemented perfectly, you know, important caveat can be fully secure. Okay so with that kind of bringing it back to Ethereum of course I guess the last slide there heavily hinted to where we want to end up at, but where were we in the past, where are we today and how could we get to this shared settlement layer future. So kind of starting with Ethereum a few years ago in the early days basically Ethereum also was a pure execution chain, we had proof of work security, roll-ups weren't yet a thing, that was kind of a happy place to be in for a couple of years. At some point of course like Ethereum always wanted to figure out like how do we actually scale to billions of users and it took quite a while, quite a few years of iterations and iterations just because I like kind of research archeology and I would encourage you if you ever have some free time and nothing else to do, go look at all research posts, it's always fun. So basically just to illustrate that a little bit, I've picked like three that kind of accompanied our way kind of towards this roll-up vision, the first one is if you look April 2019, it's really quite amazing, KCD Trio phase one and done, two as a data availability engine, nowadays I think this terminology like everyone would just nod and be like yeah sure, thanks shouting, but three and a half years ago I think there was quite visionary. Then a year later, a year and a half later we had Vitalik for the first time confusing this term roll-up centric road map where basically we finally realized that yes, there is no other way to scale, like you really need this decoupling of the execution of the settlement, otherwise like you basically, as I was saying earlier right, you can't break out of this dilemma, so really like roll-ups are the way to go here. And then just third, a pun little one that my colleague Matt wrote a couple of weeks after Vitalik's post, looking back at our early, his two plans and it actually turns out that basically this, if you remember the early shouting plans, if I just go back a couple slides if this works, actually what we were planning back in the day with all these different charts was kind of this picture all along, right, it was just that we wanted to have all these charts be like super enshrined and at some point enshrined roll-ups basically and at some point we realized it's just much better if we open this up for general innovation, if we don't build, basically if we are the only parties building roll-ups. And so yeah, this basically just you know is a little history kind of tidbit so this is kind of how we got to where we wanted to move forward with Ethereum so that kind of brings us to the kind of the pre-merge state we already started to use the Ethereum chain as a settlement chain as well but of course it still has execution so it's like kind of like a hybrid chain still proof of work security of course before the merge and as I was saying settlement chains need the settlement aspect which you kind of get for free when you have a general purpose chain but then you also need data availability and data availability on the existing Ethereum chain is really quite expensive it was never meant to provide data availability so it's kind of basically all kind of hacked in you just basically just dump data into Ethereum transactions and then never actually use them which works but it's just very very expensive as a source of data availability of course we started having roll-ups on top really exciting if you ask me I think roll-ups have been making a lot of progress over the last two years and then you know pre-merge we had this beacon chain just an empty little box nothing in it but much stronger proof of proof of stake security right and then of course as of today we ended up you know merging and now we have have this basically same situation before but much higher security and just as a maybe quicker side I think I have the time for that why does proof of stake actually give higher security I was I mean it's not really kind of part of my talk but just because I always see a little bit of confusion about that debate on Twitter I think it's really quite quite simple actually like the way I would think about moving from proof of stake in terms of security one it's just more efficient right and that is because you don't basically with mining you actually have to pay for the entire cost of the operation you have to reimburse miners for both the mining hardware and the energy like they have to make all of that up with the income whereas in proof of stake you only have to make up the lost interest on your money so basically you can get away with paying way fewer rewards for the same security in proof of stake and then the other one is also that proof of stake security is just fundamentally more effective so that means that if ever there were to be a failure which of course hopefully there will never be but if there ever were to be like an attack or something the attack is attributable and so you can go in and selectively punish people in proof of work once someone attack has 51% of the resources you basically just you know your change is just lost nothing nothing to be done in proof of stake it's really not a big deal actually right it's a little bit annoying for maybe a day and then afterwards you just done with it so that's really really neat I think and then also like I couldn't quite help myself maybe this one esoteric maybe but another quite interesting mechanism in terms of security that I really like is this ultrasound money one where basically the idea is when you have normal mining rewards that means that we have ETH as the asset has just much improved monetary properties which in the long run might you know might or might not turn into a higher expected monetary premium higher total ETH market cap and we can just buy more security in absolute terms but again you know this esoteric totally fine if you don't buy into this that the first bullet point is the more important one here okay so this is Ethereum today where do we want to go how do we actually get to this full you know share settlement chain how do we get full data availability all of that kind of thing so first thing is next year hopefully a fingers crossed looking good I have to say proto-dunk sharding EAP4844 that we've been really actively working on just earlier today there was a session on this where basically we I don't know if you saw that but like the data symbol there just become became way bigger because you just you know you add better data availability to the Ethereum chain for the first time data availability is an explicit service we provide it's not just something tucked on but it's an explicit service and so it's just more effective and then of course you've probably all heard about that dunk sharding soon TM afterwards that could be two years from now hopefully like maybe maybe it'll be three years but soon and that's really where we get like ultra scalable data and with that we will be for the basically we'll be able to to host multiple of these really high throughput roll-ups so this is kind of in terms of scalability this is really the end game that we as Ethereum basically are aiming for and basically kind of with those pictures in mind I kind of wanted to then for the for the last part talk a little bit about where does this leave us like this I hope that kind of gave like intuition wise like a good idea of where we're moving and why but but what implications does that have so the first first thing I wanted to talk about what does it mean for the existing well each one we used to call it each one chain like basically what now is just the execution chain within the beacon chain what's the future of that and they are basically multiple different potential future visions for it that you could you could see the first one would be that it basically stays this hybrid execution settlement chain all the roll-up settle on top of it but also still you know it has some defile on it whatever for high value transaction it can still be used as an execution chain so that would be kind of the default case and then we could have the scenario where it really turns over time more and more primarily into a pure settlement chain so that means that really more and more of the use activity migrates layer 2's and layer 1 is really just becoming more and more just a roll-up management layer and then the third one it's more like a fun thought experiment it's not really kind of planned or anything but just wanted to kind of illustrate how this modular architecture really lends itself to all these different ideas what we could also do is that we could basically add an additional separate pure settlement chain to the beacon chain and turn each one basically into a well you could say a roll-up right so in this case if you look at that picture you could just think of you know like one of those roll-ups being each one so that's again I don't think that's likely that we will go that way but you know it would be almost I don't want to say trivially but it would be very feasible to go that route because we now live in this in this modular world the other one I briefly want to talk about was the future of the EVM first I do want to say by no means certain that the future of high throughput execution chains will be EVM based I think Ethereum as the settlement layer will always use the EVM I think it's just good enough for that but it's not yet decided like it could be that you know Solana or something has the dominant winning highly scalable EVM solution it could be the EVM could be some of these other competing new ones this is yet to be seen I mean we are very optimistic for the EVM in that scenario as well and I think I do want to point out because that sometimes misunderstood the reason why Ethereum only has the limited throughput it has is not the EVM the limiting factor it's really that we want to keep the resource requirements for validators low the moment you are a roll up and you adopt the EVM you can still go far beyond the throughput constraints of the base layer and there are many possible EVM improvements that we just never bothered to do on the base layer because the data is not relevant for us but that layer 2s are very actively looking into so I think there's a very good chance that the EVM will turn out to be like the winning or one of the winning high throughput VMs but of course the outcome of all of this process will have big implications like clients will basically say the existing EVM clients will there be primarily settlement clients only or will they also be used for roll ups at some point maybe they want to focus on roll ups even how does it look like for programming languages for the EVM what about tooling all these kind of things so I think this is going to be one of the main interesting kind of technological debates of the next five years and in particular I wanted to briefly, very briefly talk about EVM equivalence because that's just a topic that I personally am really interested in so for these layer 2s, for these roll ups that choose to go with the EVM the question is do they go with the exact EVM or do they just go with something that looks kind of like the EVM and if you remember say optimism for example they started out with what they call the optimism virtual machine or something OVM which was basically an adapted version of the EVM and now they're back at using basically the EVM as is so the question is in the long run what is the way to go should you go with EVM equivalence or not and so just briefly basically the pros that I see would be of course standardization like if you use the EVM just all the tooling works out of the box you have multiple existing client implementations which again for security having multiple clients is just really really helpful but also really hard to achieve so this is like a big win you can get with EVM equivalence the third one is a bit more speculative but we could definitely imagine in the future that we have some sort of specialized settlement functionality around the EVM that we offer for roll ups of course that would not be mandatory if you want to run your own like fuel or Solana or something on top of Ethereum you can do that as well but we would have special support for these EVM based ones and one really interesting one especially I think in the last few months people have more noticed how governance can really be a liability and if you actually follow the EVM as is you can just defer to layer one governance and follow their changes basically and then what are the cons why might you not want that well of course one it's just if you're stuck with the layer one EVM that does come at a cost of slightly slower iteration speed we have to really vet every change that goes into the EVM quite severely to make sure nothing ever breaks you might just need some layer two specific functionality it's just harder to add that if you really don't want to mess with the EVM and also and I feel like that's the one that's the most relevant in the context of my talk here if we expect layer one and layer two to really operate at quite different scales in terms of throughput maybe having identical VMs is just not the ideal way to go and just as like a potential compromise here or having the best of both worlds idea that I want to leave you with here is one that I think might actually be quite exciting to explore in the future what if we could maybe get the best of both worlds by having a dedicated version of the EVM specs for layer twos that's still standardized across all the rollups that want to opt in but is able to over time depart from layer one and really focus on optimized for the layer 2k so I think there's a good chance that we might end up in that world and I think that might actually be the best way to get us to high scalability and yeah I think I hope if I remember correctly oh I do have a summary slide so let me briefly go through that but then we're done so you know traditional blockchains you always have the tradeoff between security and scalability rollups solve this Ethereum's vision we want to become the primary settlement chain we're still at the very beginning of this transformation Ethereum will likely this is the last point I add the likely because who knows right but we do still with the EVM also have ambitions for for execution chains we do not just want to be to be a settlement chain yeah and with that thanks