Kautuk Kundan I General Purpose vs App Specific Systems I Stackr I Pragma San Francisco 2024

[Music] and for our last talk of today we're going to bring on kotuk to tell you about how do you think about general purpose versus application specific systems and this kind of includes everything about rollups VMS stuff you're seeing in ZK stuff you're seeing about risk all that stuff is all part of this way to think about and understand this so please give katuk a massive Round of [Applause] Applause hello thanks Karthik uh hey everyone so my name is kuk and I'm the found of stacka labs uh talk more about it in the coming slides so the topic for like my talk is general purpose versus app specific systems uh I have kept it like a pretty broad uh overview of different things and then I'll go over like what's like the best fit for you so a little bit about me as I mentioned I'm the founder of stacker laabs and we are building something called as micro rups and before that um I I like doing protocol research before that I worked on Hubble L2 and BLS wallet at the EF and then I also moved the lifeb protocol to arbitro so coming back to the topic so this is like a basic example of like what general purpose is versus what something best purpose built can do so a pickup truck can do like a lot of things it can drive it can tow it can store stuff but like an excavator would excavate and it can do so that pretty well uh similarly like a more like software example would be on the web two side we have Twitter and then we have Tik Tok so the Twitter you can have things like text images photos videos audios uh whatever you can think about posting on regular social media and then on Tik Tok you have short videos which one is better that's a very subjective thing I don't think anyone is better than the other it's just that one serves the purpose for which it's built for and then more uh Crypton native or like a web three example would be arbitrum versus dydx so arbitrum is more of like a general purpose chain you can or any kind of application on it whereas dydx is a more closed system wherein it has been optimized pretty well for doing trading so again coming to the definition like general purpose uh it does a bunch of things all at once whereas app specific systems are purpose built for very discrete use cases and sometimes app specific can be loose uh and sometimes can be very strict uh then then you can split app specific into domain specific it's going to come a little bit later so is one better than the other so that's like the question that we are trying to answer so let's talk about it one by one so coming on to this famous code which jack of all trades so general purpose uh systems the features of them are like they have substantially lower entry barrier because everyone uses it um you know how it works there is a lot of tooling and like um support for it available uh on different forums so you get familiar tooling you have a lot of like people building on it already and it can also act as a base for many app specific use cases so it's not just like you are limited to that General purose system you can have have it as a base for other systems as well so some of the examples in terms of chains are like ethereum Solana like you can have like ZK VMS which are like sp1 Risk zero and there are other examples as well so very very uh like uh it's going to be a reach but there are some specific cryptography as well so I think Remy talked about F that's like a application specific cryptography wherein you only care about like encrypted data whereas later there general purpose cryptography like ecdsa signature verifications like everywhere but Jacka fall Trad master of none is something which uh we have all heard about so there are problems with it so general purpose system so ethereum has scaling issues we have all been working around it so we have been building rollups different solutions for it salana has decentralization issues because of the system requirements that it has it has it's like a very high system requirement you uhu lose some of the decentralization because of it but then again like it does that uh thing uh does the pal transaction very well and with zkm they are not too efficient so we have been like recently benchmarking a lot of different zkv Ms like the general purpose ones that Target risk uh risk five architecture uh you can run a bunch of compute on that but they are not efficient so if you are doing something very specific you are bound to write your own zkm from sketch or like just uh write like circuits which are able to do those kind of things that you looking for so these are the specific issues for us a general purpose that just does too much that is not uh generally for scale so you can build applications which goes up to a certain extent which is like the limit of that particular system and and then you always have some compromises attached to it so the compromises could be in terms of compute it could be in terms of uh uh Noisy Neighbor problem you can be limited by a set of different things that you can or cannot do with it so how about bespoke system so I talked a lot of things about uh General system so am I trying to get in beef with all uh all the systems that are out there so is BB systems really the best so some of the features of our speciic systems are that in they're infinitely scalable and when I me uh when I write infinitely it means that you can choose to do whatever you want with it to a certain extent uh like in the scope of your application or something like on that and they're useful scaling because uh if you're building something very uh purpose built for your use case you can just customize it to the extent that it can go to and definitely there's a lot of uh room for Innovation so for example like if you guys know about uh Formula 1 ring most of the advancement in uh car technology has been coming from like those racing U advances so all the aerodynamic stuff all the engine performance all of those things Formula 1 is a very application specific thing but then that rolls down to general purpose things like regular everyday cars so some of the examples are like chains which are purpose built so csti is a chain that is purpose built for doing data availability I mentioned about dydx and then there are other chains as well which are application specific for uh yeah for different purposes then for example of zkm so there's miden which does uh private private uh computation on like uh it's an L2 that does like private uh transactions then there's delines which is a wasm based zkm and it can run uh wasm very well if you try to run uh wasam on a regular zkm it's just not going to work and then there are dsls like Cairo or Noir or like some other languages as well so these are like purpose built for doing uh cryptography and those kind of things and obviously there are a lot more examples around that yeah but at the same time even app specificity has a lot of like issues with them so to develop uh it's a little bit difficult to develop on DSL so you have to have like a very good understanding of the system itself to understand what's happening how things work uh how to resolve bugs and all of those kind of things bootstrapping applications as well e as well as ecosystems is hard because to build a particular application for a particular system you need to understand the nuances of the system you need to understand the nuances of the ecosystem that's going to like that particular system and similarly since something is very uh small it creates a niche so there's a lack of community support for them uh in general you don't find a lot of good resources around it because someone has not built a YouTube tutorial trying to talk about it and in general they just get disconnected so one app specific system can't really talk to another app specific system u i either because they are like two distinct systems or they just don't have like the right set of apis to communicate so that creates like silos so again like coming to the issues so they create silos and niches so silos in the sense that these uh systems can't really talk to each other niches in the sense that there are very limited set of people who know what to do uh this is very resource intensive because you need to find the right set of people you need to find the right set of infrastructure requirement to run these systems and in general like it's it becomes like eventually hard to do that and it is very easy to get into the analysis paralysis uh scenario wherein you can keep on customizing this over and over and over again and you just don't know when to stop this something which we have been facing recently as well so that's like a few issues with app specificity as well so now just to complete uh the code so Jack of all trades master of none and it goes like uh through often times better than a master of one and this is something which I feel uh that's going to be like that that evm sucks but unfortunately it's not going anywhere because all the new chains that are launching are going to uh are choosing to do uh go with evm so what's the solution so is there something that exists like which is picking the best from all the things uh Best of Both Worlds like picking stuff from the application specific side of things picking stuff from the general purpose side of things and just combining them together I would say no but we can average them out together so there are good parts of some certain systems there are bad parts of some some certain systems but what we can actually do is just combine the minimal averages of them and get something which is good enough for most purposes so that's something uh which I feel comes into the category of domain specific chains this is kind of like a new term I haven't seen a lot of people using it but this is something which I feel fits perfectly between a general purpose system and an application uh application specific system which is domain specific chains or like domain specific systems to be honest so what is it all about so what's the definition so if you go about defining domain specificity it's like basically taking a general purpose system and restricting it to a certain use cases so it's not that you uh strip away all the important bits from it you uh just customize it uh completely but it's that that okay there's a general prop system you modify some parts of it which are relevant to you or something which can make things better until you get to like a good enough system so the benefits of it is that it creates a low entry barrier because you already are familiar with the toolings available for it there is good community support and there are a lot of people working on those system but at the same time you can customize it to like a really good extent because uh you know the system there is tooling you have worked on it and it does not feel disconnected because if you have like a minimal set of requirements you can have other applications uh that can communicate with each other on that particular General uh domain specific system so what are the examples so most of the evm based custom l2s fall into this category so recently we have uni chain sonum launched Kinto which is like a evm based system but with a uh kyc enabl so all of these are like domain specific chains uh and most of the chains that are built using op stack fall into this category that they give you an evm uh evm based system that you are familiar with you can deploy your contracts on top of it which you have already written and then you can have more customization on top of it by changing the block Time by changing block parameters and all of those kind of things so what's the end game is there an end game so I would say that the end game is nothing but uh app specificity with good tooling and user experience so you can get to a point where in the app specific things don't feel very disconnected you create a good ecosystem around it and then you create a good set of tooling around it and the community just comes together and builds something good so this way the apps don't feel disconnected you still have like good uh amount of things going on it's efficient it's customizable and you can do whatever you want with it and the good news is that app specific or like domain specific chains are on the rise so if you go onto like the selenium Explorer you find that uh they list down like the most uh uh active chains on their ecosystem and out of the 20 there just two of them which are like General general purpose but most of them are dedicated to a particular use case so it could be either trading it could be either uh kyc verification verified transaction and all of those kind of things but uh yeah that's like something which Isen happening these days so coming on to the major part like how to decide what's the best for you and to get there you need to answer or like you need to ask yourself these three questions what tech features would Li would limit your application like is there something which uh you need to care about which the current Tech uh which the current like chain does not provide you or you need to maybe customize the VM a lot so that you can you are able to do these kind of things so for example it could be like the block times like you're not happy with u 12C block time you need much faster blocks so that is like a limiting feature so you need to be more bespoke for that how much resource you want to spend on it so that includes not just the resource you uh spend to build that system but also to run that system so in case you are spending some time or like money to get some developers to build on that particular Tech feature then you need to count for that and then also it's not cheap to run a app specific system you need to think about the infrastructure cost you need to think about all the other things that goes around it so I think hilber from like one of the previous talks mention how cheap it's going to get but that's something that you need to consider and I think the one final thing is that how do you want to capture value so capture value mechanism is like is that something that is enshrined in your protocol you take fees or is that sequence or Revenue that you are after and all those kind of things so yeah this is one of the things that I've found on Twitter so it has like a really good mind map on how to think about launching an application so let's say you have a Dap in mind and you don't know what you're doing it's something that you really want to experiment with so you just build build it on a shared chain see how the community receives it and then once that is getting a good product Market fit think about launching it to multiple chains so once you do that or you maybe have maybe you're just taking a proven use case and you want to do that so just build across multiple chains make it a multi-chain application but at the same time you hit the limits that okay your application is so popular that it cannot keep up with the demands of the system then you go something more bespoke and then you build an app chain uh that can connect with all the other chains so connecting is important so it doesn't need need to be an asyo and to build an app chain then again you go back to the requirements on what kind of tech features do you want what kind of customizations do you want uh there is one more uh chart that I found uh which was by John Sharp uh so it's basically the app uh versus app chain cycle and this is something which is really true and has been happening in the space so starting from like the yellow box apps on general purpose chain wants more customizability sovereignity and value capture so they end up launching their own app chain and then the app chain gets a lot of users uh lot of activity and wants more value capture so they go about having more players in the ecosystem to build across them and then the app chain becomes a general purpose chain and then the cycle repeats one of the application on that particular general purpose chain gets big and it keeps going and going and going so that's something which you need to be aware of as well and this is something which I uh think recently is happening with AO so AO which is like a app specific chain is thinking about opening its uh chain to have more applications built on top of it uh so yeah and this is where uh stacker comes in as well so stacker SDK lets you build your own app specific l2s it helps you decide the three questions that you ask uh that I that I talked about initially and then you can uh have it uh your way so stacker is nothing but like a minimal rollup framework uh it gives you uh tooling to launch your own l2s um you can customize the state structure you can customize the VM structure it's not like it doesn't bound you to anything like that but yeah uh keep in mind like uh the kind of designed decisions that you make it's very customizable but it's really easy to get into like the analysis paralysis so however often time the reason of launching L2 is widely different so it kind of sometimes goes beyond the three questions uh initially and something which I asked on a telegram chat and the answer was this so you just need to capture money uh and how you do that there are like three ways to go about it uh two ways to go about it and it's actually true because let's be honest most of the lach most of the chains that are launching as separate l2s want to capture the L2 uh want to capture the sequence of fees and this is just the data from past year from uh from Dune and you would see that most of the revenue is upwards of uh $10 million and it was like a crazy Spike back in March and then it has been like sustaining all of the U across the across the different months so yeah so fee capture is another thing that you can think about so one of the ways of doing it is running your own sequencer but with running your own sequencer comes in the issue of decent ization so the applications are just stuck with that uh it's a overhead because again you need to pay more to run the sequence and all those kind of things and then L2 bit won't like you because then it comes into a different category of stage zero stage one and stage two rollups and there is something new which is coming up recently uh it's like a very weird name I don't like to use it but it's called application control sequencing so up until now the web has been uh uh a structure wherein the protocol layer has very less value caps and the application layer has like the most value capture that has this has been like a lot different in the blockchain scenario where in the protocol layer itself has like the maximum value capture and the applications deploying on top of it is not like capturing a lot of value and this has been like the most uh this has been like the biggest reason of discomfort with the application and then they go about launching their own protocol because the protocol capture values and then the application would eventually get value from that kind of thing and this is something called as a fat app thesis wherein uh you get like the application layer itself to be fatter than the protocol so that that that's where the value capture happens and this is uh this is another uh thing wherein you don't force an app chain you the app chain should come naturally and app chains are definitely not the end game but you should you can think about like the different tradeoffs so launching an app chain it has a very high cost but you also capture high value and that's some some decisions that links back to the three questions and then chains that share value they are definitely like lower cost but they are mid value as well because you you can have domain specific chains wherein like let's say there is a chain with most of the defy activity happening you launch your application but then the other applications on that chain would uh probably Share value and those kind of things so this is U an an example of that could be blast like wherein like the sequencers commit to an application but then again you don't capture the 100% value of out of it and then there is app specific sequencing which is like the new thing coming up it is midc cost but the value capture is something which we'll yet uh which is yet to see uh we'll be seeing that in the coming days uh but yeah this sounds like like a promising thing that's coming up uh yeah so as an advice as an infra developer is that you don't need to build more infra uh what you need to be focusing on is doing more of the app building and let us take care of you and uh yeah that's that's about it thank you [Applause] [Music]