Hashing It Out
Hashing It Out

Episode 67 · 1 year ago

Hashing It Out #67-Optimistic Rollups - John Adler

ABOUT THIS EPISODE

Layer 2 scaling solutions have gone through a lot of iterations and disappointing conclusions in the past few years. There is a newer solution, which is boasting some positive (and safe) results: Optimistic Rollups. John Adler, Blockchain Researcher at Consensys, speaks on the history of Layer 2 solutions and describes Optimistic Rollups and how they may help with blockchain scaling.

Links:

Owinjurywelcome to hashing it outpocast for Resolk to the tech, intevators behind blocked inintrastructure and decentralized networds. We dive into the weedsfigured at Whyan how people build this technology. The problems they facealong the way come, listen and learn from the best in the business. Yo canjoin their ranks all right, Wolk EM, back to hashing itout, has always on your hose dor cort, petty with my cohost calling cuchet. SoWat's up everybody gone. What's up everybody calling and todays episode,we have John Adler, who is going to tell us all about the roll up and thedifferentiation between them and whatever else he feels like talkingabout John welcomed o. He show why don't you sart off by doing the normalthing and telling us kind of where you're from how you got to the spaceand what you do? Hey Cor? Heycolin, it's a great to behere. I guess I'll start off with a briefintroduction to myself. So I'm John Adler, I'm currently with consensus, doing etheriumscalability research. My background, I have a masters inelectrical and computer engineering from the University of Toronto. It wasmostly focussed on formal verification, any baggain of hardwaise circuitsfairly different than you know the crypto space, although the FORMLvarification does come in handy when there's l large overlap, Etween, u knowhard ware and software, so it comes in handy when I'm dealing with projects, Iclaim through forl th verification of Smar smart contracts. I also had a pretty deep interest that,unfortunately, I never did any formal research on. I wasn't part of you knowmy my P, my graduate studies H in compilers and interpreters, and ithappened that this had a very large. I saw that there was a large overlapbetween compilers and interpreters and what people do there, how they optimizethese things, and you know the raplicateere state machines that areblockcans and things like you know: How do you desigre smart contract language?HOW IS IT efficient? You know what's the performance of of your Blocksha and so on, and those aa Wuge amount of Orlo between these two h from what I saw at the time- and Istill think Thas true and that kind of wadrove me initially to kind of entureand be interested in the block can pace which was maybe about three years agoand then about a year and a half ago I was done with my studies. I put my Iwas in. I was doing a PhD and then I put that in a hole to join consensus,to do layor to scalability research and it started off with Plazma UH fairlyearly on. I kind of did an overview and saw thaut a lot of the plasmobarianshad a number of issues with them. The primary issue and a I mean, there's abunch of issues, but the primary one that I saw was that they were allpermissions, I'm kind of in thing around the time around a year and ahalf ago or so was people were saying. Okay, we can have a system that cansteal er funds, but it's permission and that's the Tradeolfi make and I didn'tfind that trey off acceptable. I wanted something that was both missionless andtrustless andscalable and that led eventually to them SI reaps. So sincethen, I've been doing later to scale ability, research in the R context andMor recently I've also been doing work on Phas to afitherium too. That's my background. I'd say: You'vegot your hair work cut out for you from here Doyou. So, like you keep hearing this,that was a big. I guess you know, headliner of the recent dve gone wasyou know, th the next greatest thing for scalability in terms of the theoryt toois like it's no longer plazmo, it's this thing called roll up. Itstarted off with a Nozila Age, roll ups and then okay. Well, maybe we're otgoing to use those we're going. Ta Use optimistic, Rols instead 'cause theyre,better in some way, I think, there's a lot of words. Tho have thrown aroundhere that people are saying cool, but they don't understand the differencesright. How does firstlike this and start off with like what ies theconcept of rule up me, especially in comparison to what plasm like what mymental model of plasma is and thef Yo can talk about the varians from there? That's good. How about? Do you mind ifI start instead o instead going over some of the more traditional andhistorical scaleability proposals, things are channels and Sijans andPlazma aneour audience does kind of have a background like yours. So forthe most part we do kind of understand an we do go over this in detail inother shows, but yeah like like tell us what built up to roll ups 'cause, I'mreally clike. I followed plasma pretty closely and then I realized it wasn'tquite going to work the way they advertise and then I kindof likechecked out. It's like bringing it back in o a reat yeah. So I'll go over someof the challenges I mean. I, since you...

...said Your viewrs are more to O. that'sgreat. That means I don't have to go over everything and you know metaculorsdetail. I can kind of assume they know how it works at the base and then kindof go over some of the challenges and, like you said, this'll lead eventuallyto Wy. Wyoll up is doing what it's doing so they start with kind of allthis ones, which is, I don't know if there's oldest ones, 'cause. You knowthese is all Tey D, All day back to bitcintalk threads sothe th, no one.That's traditionally usein the bit goin space, which is payment channels orstate channels. You know systems are lighting, atworkrate and connect and somone theres, an bunch of stat channel projects, and it's you law coins between two ARD,more parties and then the state progresses through unanimous consent.So all the parties need to sign a state update the simplest case or thesimplest way of visualizing. It is that Nei everyone just increments, a nont,so every time they sat updates, then theres, none increments! That's how youkind of version. You know different states- and this is really great. So I'm not soto be clear. I'm not saying channels are terrible to have great propertiesthat block Chans can't give you they have absolutely instant finality assoon as the state progresses. Everyone knows. That's the latest tate andthey're, basically free between the participants. This is you know.Absolutely amazing. Bloc chands cannot give you thise properties. If you needthese properties, then no channels are good for you, but they have certainchallenges and certain issues they're capital anefficient, because theyrequire full precollateralization. So all the party need to lock up all thecoins before using the channel, and this is very capital, inefficient andKno capital Crossmonke, the QUITEDA costmonke. It requires constant user mondoringmonitoring because you are responsible for your own channel or you have to payfor a watch tower. Is it also a vulnerable to traincongestion? And this is Spati Thei case if the block capacity, if the mainchain is very small or it's intentionallylimited right? If thechain is congested, people can start closing channnels fraudulently, andthen you won't be not old, honest partieas will be able to respond totheir challenges within the timeup period and there's an inbond capacity problems.So you can't, if you have something like a payment channel, tnat work, youcan' receive funds unless you've already spent funds h and then, let'ssee what else, there's a fix tat of users and the fixed thed of users meansyou can't build things like unisop on it. You can build open protocols h, let's see yes and that's basically I so these arekind of the challenges or oun channels. CS No ALWA, say: okay, the last one wetalkd about is you know the open USS. That channels are a fixed out of users.I mean an the collate equirements and all that stuff. So do we have a systemthat doesn't have these particular issuesthat we can in fact build Juni Sop on it that Ou know anyone can join andparticipate in any time and the answer wis? Yes, we have sidejins sor examples.Examples of this are liquid rithstock and scale. I sometimes call these unqualified signcrains, not because they're bad, but because they have w no furtherintrinsant qualities to them. They're just playing side, cincs whid, youconsider, like AOA network, also side chain, fors, yeah, yeah h. So thesesizans are great compared to channels because they've opened a participation,so you can build things at Unio, stop on them and they have noprecolateralization they're relatively capital efficiens, and you know, t the general the generalway that work is. You have someset of bloc producers Thak, you APPROVEFORK,grostic confederation, some authority. However, you want it to secure theSidechan. This is a completely different securitymodel done, for instance, channels or you know the main chain, because nowyou have trust, you have to trust that the H wh these Bock produsers, that aresecuring in the side chain, wants produce a an Invali block or produce aninvolve blockand withold the data, and then you know just steal all your funds.So there's a fundamentally different security assumption h and this kind of manifests itself-and this particular point is important, because this is the sanction to meanplasma and unqualified side chance. Is that there's a two way pack? This is kind ofwhere the issue ot side Jensis is the Tuapeg, the forwrd peg from the Ma Ento,the Si ense reduces deposit coins into a contract, thet getnmented on thesidechent, the reverse pegs are going from the Sijan to the Mainchain, that'sthe one that is. This is where all the difficulty lies right, because, sincethe security of the Ajam is lower, someone c can create an Invaldad blockthat just gives themselves all the funds on the Sideshine withwhole daddata and then provided like an SBV or alight client proof back to the main chain to unlock all the funds in thecontract on the mainchain and steal all these funds. So there's a fun amount,fundamentally different secutive model, and this kind of how can we do thisriverse pack without having to trust the sidejin operators or the sageantblock? Producers is what plasma attempts to do so, there's no clear definition, ALplasma to the day, which is unfortunate,...

...and I don't think we ever will, but thegeneral way it's usually like favor old mindset. F, you talked about Calvonabout this a couple of evisodes ago and it was like this is the general conceptof Plazma Braz, there's no real specific specification to what it means. Yes, I ttene to the define it in someways, but ou know the PLA, my guys kind of pushed back and they kind of gavetheir own bage definitions h. We can talk about that later, but who no colardefinition generally or the way Georgi dsescribes at which thing is a gooddescription. Is that it a side Chan, that's safe, underdata unavailability,so o r coins can't get son, even if the data is withoult even of the blockpproducers on on the sidechain withald data. So the way plas work is that you have asingle central operator and the reason you can't make PLA o permission less isbecause of data withholding in the PERMISSIONE system. The majority couldjust withol data from the minority and then Escentialy just kicked them off,because then the minority conproduce te blocks, they don't have the TATA, soplazma is permissiont. There's jeest no way around it. So we have a singleoperator and they commit to asige and block Hashes Onchan. This is nicebecause now they kind of have the same irreversibility of the same security asthe main chain. You no longer have you know. Septat consendis brought a call and deposits now the work, the sameWysidechan is that DOS lock funds intot into the contract and they convince iton the sidehand and the reverse. PAG is kind of where PLAZMA has to do all thework. All this complact, all thet COMLAC Flasma, is around the revespect, so it's maintained by an exit game sofor Plasma Nvppariance, which were kind of the original ones. You know theminimal viable Asma, more Bible, Zm o those kind of things, Yeahabe convisions as quite fun, for alittle while. No the exit game is basically just a cute and if a block iswithheld by the operator, everyone is supposed to exit their state from theprevious block to whateveris held or you know. If they have an invol block,they you can a smit a frod proof, but if oetter with withholds a block,everyone is supposed to exit their states from the previous block, whichis assume to be vallid anow. The issue is: There's a very kind of small windowof opportunity for them to do this. They have two weeks or something and ifwo think about it, they now have to exit or dump the entire side chainedstate, not just a single block, but an entire. The entire state, which isbasically unbounded in the size on the man chain, which caused a huge amountof gas within two weeks. So in other words, the kind of safe throuhput offall PLASM, N B, p chains running on a single Manchain, not just of one but ofall of them put together, is uper bounded by the free block spase of themainchain. So what's the freeblock base on a theorium right now, approximatelyI don't know if the top of my head, like approximately therall right, Holl,ust, ack Acabali, is being used. You yeah exactly so so on on ithe th, the te safe through put of all PASMYBP Trans is basically zero. That's why plasmame?He is basically that end. Everyone gave up in it and there's a number ofprojects like o Musigo, an madic and soon that are still working on this,because I don't think they understand that this is the t movie, though yousaing that we're not at maxcapacity on every block,though I've seen blocks that er some blocks or low and some blocks are high,like I mean e almost may issue there, even if it's even if that's the case,it's it's variable and draticl. It is very O drafticl Limis,the CIZER Platan chank and have which is not a scaling solution right. It'syeah 'cause, it's over two weeks, right so shure each block in obariancs, butover two weeks you say: what's them ot a fespace there and that's you knowit's basically thr over oirers, you see what, on average, the freezemace overtwo weeks would be yeah that would respon O exaen I on five percent. All Ihad to say from what I've seen is. Probably like ninety yea, but if youwant that, a as as a general framework for scaling out, that's not going towork because as that popularity increases that, like like the security model of like what,if both chanse exit but the size of Ales thing, Ai Ysfer all of thavailable chains is, it is, is bss much but smaller as gross, and so that's notnot a scale. Essoltion yeah, that's exactly yes, so plasmp was is Adenoone,no one. That knows what they're doing use anymore and then everyone movedover to Plosm cash parience. So the posmcashvariants use a coin data modeland it's based off or inspired by Greg Mashel's coin, witness, which is almostexactly the same thing as Pasmacash julihely change change in Hel. Theoperator works H and I kind of liken them Tho channels, so pretend that youhave a channel between two parties. You know two parties lock up their coinsand then one of the possible state updates you can make. Is You delegateyour signing to another party its? As and Bob? U Know CR rat multi sticktogether the deposit, their coins and then one of the state uptays they makein the state channel. Is they allow Charlie to sign on behalf of AliceRight? So now really this channel is between Charlie and Bob.

Now this usually should never be done witha channel, and the reason being is that you can assentialy double spend thisstate update by signing with the same nons and everything you can sign. Astate update, IGOSAND Bobinsin state uptdate. That allows, for example, Daveto asign on behalf of Bob. So now Charlie has you know: Stat assigns date,update that says hey. He can sign him half of Alice right going forward inthe future and day of his one. That says he can sign up Bahalf a bob andthey both have the same nons, so Sanci, Alizan, boher kind of double spent theistate updates. If you wanto call that and how do we resolve double spans main chame, like bloolong as I ate the block Chai Ye,the block Chines, so you resolve double pans by a blockchand. You order themright if you orderd tenactions and there's no double spending. So what DOSpassmy cash do it orders these a Sien assigning or delegating off signinglights to this channel? That's what it does. 'cause 'ause, usually itll be unsafe.To do so. 'cause you can tell is ma, but if you order them- and you have-you know the chain that everyone can see now sudden th y you're good to go, youhave an ordering people can't double spand a signing of n the O Pareso Lawould be the double states, just invalidated 'cause. It's already beenspent yeah, so it's kind of like as and Bob Have you know, coin and the APLASMAcash, and then they sign it over now, like it's Bob and Charle kind of so now,Alice doesn't have the coin. Charly does right, and this is kind of how youtransfer ownership of coins and plasmacash. Using this, you know, blockchant order, a signing of channel owners which usually you cant, do in enormal channel. So it's kind of a combination of channels and and a blockchain and unfortunately, it kind of inherits the both of both worlds, theboth the worst of both worlds, not the bast of both worlds. So En Pasma cash, you well the benefitsofpomcash, O plasmp is there's no more mass axcess, just Lik chonnels. Youknow all the channels that are open on the mainchane. You know all the coinsat whov deposited. If someone tries to close a channel right, they have toprovide prove that they are the owner and if they're, not the like late,current and latest owner a later owner can then say: Hey No. I am actually thecurrent owner and I'm a later owner than you that's kind of the general mechanism.So you void Massa exits, but you introduce some other callenges. Youcan't use watch towers, so users need to be online every whatever two weeksor whatever, which is you k now pretty huge burdenunusers. It has a multiround interactive challenge, and this isvulnerab like it's even more vulnerable to chain congestion than you know,usual channels, anddror challenge. It's an onchain of joint of challenge,es one of the Egson Plasmacash, which is needed you can avoid. This issomeone initially well, someone tries to acat someone can challenge and say:Hey you, you aren't the current owner right.Please provide me proof that you own this coin and then the real owner issupposed to respond to the challenge and say hey. I do in fact on thesecontents, I'm actually doing a valid exit, Isi, its kind of Ha just so dores aninteractive challenge, and this I unavoidable for plasma cash who cantavoid this, and this also means that attackers can kind of exit coins that they don't own and thenthe honest users won't kind of be able to challenge all these things within you now, the two week period orwhatever 'cause this vulnerable thee, Tanng, congest him and then also it uses coins Ita Modeland it needs to use a coinsatamodel at CASHOIATA MO made? U Tx, so not exactly it's! It's like the channel,like I said, it'sn, something similar to channels how you you have like youknow in the in the lightning or whatever or in Radon, or something youopen a channel on chain. You know that channel is between two parties, and youknow what the funds are. The on chain knows that the channel exists it aslike a unique ID hive. You want to call it got it. This is kind of the samearchitecture that pasmash uses just likeaas along a channel as opposedto making noit. Yes, okay, H, yes, but then you can also theposit. If you wantto make non ones right, you can alt sor of Hause n new coins on you can clotheOkin withraw coins, which is quit on of closing theie channels h. So you needto use this coinsata mot. You can't use a regular UTX ODATO model. You can'tuse UTX Ol da Tamoal, and you can't use on account state a model if you need touse the coin Dhar to model, and this is unfortunate because the coins datamodelrestricts e, certain things and certain performance, so you can't have uniop on Pasmacash, forexample, you need something like an Accoun Sdatmodel if you' want o Uni Lot, so you can t do that UH and Ronside Chans. You can do UNIAright. So now, we've kind of regrassed because again pasm cash is like someweaird, some weird mix, Higbrid Mongrel, child, the founchannels and sidechins,and it doesn't have the bast of both worlds. So now you've regrassed in nterms of on peaters. So you can't D, You CAn'tbilnio slap H, there's alsovarious concerns like...

...the operator, it's very expensive tokind of mercalize the state. In this way, every block and pass around proofsis very bad with expensive and it kind of limits how many transactions thatspend the same coins can be done in a single block ands on and of course, the worst partis that as permissioned are these side? Trains are permission less potentiallyright, the viewis proof of work or Approvas tey. You can make Si changsthat are permissionless, maybe not trust less, but o these permissionlestso Pama. You know as all ot the complexity as all these downsides aspermissiont. So it's not exactly the ideal solution, that's about it for the Backos to bethe overview of other implementations of level totechnology. Yes, that's basically that kind of an overview in the challenges,and what we want is something that's scalable trust minimized so close toTrustlus, theehas, a trst, madnimize, toapeg and permissionless. We want allthrough of these properties and optimistic rowups is kind of the firstpractical design that a stually give the tist in the history of Blocktam andthat's O tha. I Wan to ask you something: Um Yo keeps talking aboutpermissionlest, but permissionless is not necessarily something so.Permission is often just an optimization around your civilmechanism. There's no whati. What are they doing?That is making it so that they have to use a classical permission foprotocoland they can't use a different civil mechanism and t a more open protocols like whatis what is actually the barrier to being permissionless in, let's sayplass by cash like what is what is h? What is that? What's o? I don'tunderstand because iike yeah so for plasma in particular, like there's nobarry or to permission less than less inside Chans righ like you, can justhave any contensive protucol you want in Sidechane for Plas in particular thereason it can't be permissionless, while still you know being safe, underdataonavailability. Is that the majority of s, let's call them Balidators r blockproducers right. The majority of the Sargan validators can withhold datafrom the minority, and then the minority can no longer produce blocksbecause they don't know what the state is. They don't know what the newtronsaction they don't know anything, so they can't produce new blocks andthey eventually just got kicked out O consendsive fo the call themierentally. The majority can keepdoing. This. Keep doing this keep doing this and just the system kind of tendsto word becoming formissionless or ten words becoming permission, even if youstarted off or seated as seemingly permissionless. What could sets thisprotocol? Were they using four four puzle cash single operator, o PoA mm,okay, okay yeah, that makes lolt of sense, see t aiatten, th o could use other thing like theet,H e. Why do they choose to do that? I feel like there's better options foreven even like multioperator POA. I mean I guess you could do like Ike ALD, B, fd, protoquafter accosswould av work like you couldve. You could have done this in a lot ofdifferent ways and that would have th that would have that would have beenfine, especially giving the small node size of these things that I mean yeah.You would require media an initial set up, but yeah you can get it going Um. I don't see why that was. I think thatwas like an optimization for development. was that actuallysomething they planned in like a long term solution ore? Was that necessary?I mean some teams are saying we're going to use proofs. They cand solveall our problems, because you know they don't understand that this is a problemand that using prove stake doesn't magically. You know make it actuallypermissionless and actually trust us, but in general I guess is because theimplementation and plasmajist didn't get very far. Ou know. Clients are only mostly halfumplemented and then usually things like Javascript withober Aso, sincethey don get very far, I mean they're, not GOINGTO, build Hea' get ome thatproblem, yet it seems so let'slet' say let's say Um. All of that is is well done, but no onehas any qualms. I'm sure people will have qualps ot, maybe particular partsof that Ut. That's not bi doell es Um. What is optimistical roll up and had itto compare to all these things? I'm very curious about this, like I'm verycurious on how this had a mentally model. Optimistic Roll up verses, thesepuegios ideas of state channels that I already have vetamodels for great. So I'll do exactly that so againto kind of set the stage which is we want a solution, ski and Stlution-that's scalable and has a trust, menimized toapeg and it'spermissionless, but trust minimize. I mean one: That's has trustless as wecan get it. For example, if you're using fraud proofs, you always have asynchrony assumption that you can submit a fraudproof that a block isinvalid within a certain pyramedlizable period of time so two weeks.

So this isn't completely trustless,because the miners could decide to censor this froud proof transaction andthen steal funds by colluding with the sidechain block producers, and theycould do this. But it's very unrealistic, at leastaccording to certain groups of people such as the you know, with thedevelopers and community that minoris doing this, like actually activecensorship for two weeks is going to go. UNNOTIC and nresponded to the communityand the bit coiners are in a certain way the same way, even if they won'tadmit it when Binance got hacked, and then there were suggestions that hSCAESA. Could you know dribe the minors essentially to do a Reorg? The BISOcommunity was up in arms about this and they were vehemently h against us,despite the fact that economically Wel, why not it? I mean whin Er's good, do tthis econom cause, there's nothing stopping them technologically oreconomically hl, Hav, inpact, Bein profitable for them to do this. Theydidn't do this because of social reasons, and you K, ow socially. We will notaccept a two week, long Rearg or two week, long censorship attack by blockproducers I'd if they did this in Bitcoin, the the MINERSW s get forkedoff and all the races would get bricked. We would you know, change the proof ofwork, algoritm or something yeah. I see some serious burning bridges thathappene exactly so socially does not happend but corners kind of don't liketalking about this, but this is this is how loctans work is not governed bymath as governed by people socially and Um in Heterin space. Well,unfortunately, I her in pace since we're using a GGP friend, the hashingalgoritm. We don't really have an neir course. If the minors decide to startcensoring transactions, whe're kind of screwed, that's one more reason thatatherium should move to something: that's a friendly, for example, catcackor Shathree or whatever you want to call it or something else. THAT'S OAIfriendly! That's not my problem, but regardless we have. You know social,certain social expectations that such a Reyork ore such a long censorshipattack, will not happen, and but it' it'll be incorrect to call themcompletely trust us. So I prefer the term trust minimized. So we wantscareboll trustminimos to a Perg and permissionless. We want these threeproperties, and this is where Optimisticgra, sometime so kind of the roll up idea in general,which II B history about was smarted by the role by Thein General, which illgive a bief history off was started by Barry White Hat sometime in I guess,the summer of last year, Dont Canowday, some terrible tats. He started arepository called roll undersore up which describe nor scheme or usezeroknlege proofs, and then you have data onchain or whatever, and you do the traditional wold ofstuff- and this was later improved by the Talabuteran. In his fairly famouspost, know, Theit scaling two five hundred tren Aton per a second on eathresearch. There's, I think the second most visited post on eagth research. Idon't remember what the first one is, but it was kind of not not relaventtoanything. It was like some, some governance, post or something but oknow this post is ection h. You know extremely widely viewed andtalked about in this cast h, so risn'nt a star wit, very white, Hart wit, TelPtreand oproved it- and you know there were various talks about using watchingas a data bile bittylar. The assumption was always we need to use DK proofs. Weneed to use this exotic new cryptomagic bat. No one really knows how it works.T IES a bunch of weird assumptions: It recards their trouses that up itrequires you know extremely expensive computers to to generatly, proveECCANRA, etcetera. Theres sometion was always that this was kind of anecessary component of these rereall of type structors now and Thos, and the reason I wouldventure to guess that historically, this was the assumption. Is that Ou n?This is how people are pitching. It is that using zice, the tezike proofs, youdon't have to do verification on chain, because you just you know, verify avalidity proof, and then you don't have to verify all of the transactions. Soyou kind of have one person do the proving and then the chain can do theSinatification, and this is where kind of the scale ofdo scale abilitybenefits come from. It turns out that this is actually not really true andthat's not at all Wa. The scale B en ESCA building benefits coing from whatreally what really comes tro him is that the way robs works is that youpost, I guess I'll star with e offs, which is that Yo Ge Ropsis, you post AlAll the sidechain transaction on Chan like every time, theres new, Saan blockyou just post entire setup, toactions on Chan, and then you kind of marcalizethem or c Hash them. You passd this to, and then you also smit a proof that these transactions are Valis. Dataansission from the previous sdate to the next tate and the previous date, orthe kind of the current late estate is all iy say by contract on Chin rightand then you verify the proof, and you say: Okay, do these transactions ifapplied to the current later state? Do they in fact give this next tate?...

And if the answer is yes and everythingverifies, then you save that next dat? As you know, the current latest StateOntrin and essentially the only thing e could keep on chain in storage in nearthe Entran state. Is a Mircel route, a state rout off the side, Jan State, andthis is actually where the scale ability comes from- is that you justpoles the transactions on chain. You don't actually have to verify them theverifying that they are valed through a validity. Proof is one way:ofvalidating the transactions, but they're in fact, other ways, which isthat you can use fraud, proofs and interactive varification games insteadplus es sincony assumption, and that also has the same effect of Valadadalatating their transactions. So you don't actually need the zero inelligproof this. This does not actually confer scal ability benefits. The skillBil scall ability benefits are because transactions are just posted on chainand they're, not validated on Chan, because, for example, in Betlin H,let's consider Betcoin h the two most expensive things in betcorn issigniture parification y. by far, and eventually it's going to be state, theUTEX AL set is, as time goes on, is going to keep growing and growing andgrowing, just like the state in asyrium is going to keep growing and growingand growing and with no, you know, staterent orwhatever, to mitigate this. Every time you add something to the state. Itmakes all ather state operations that much more expensive and it's just goingto keep getting more and more expensive. And, unlike history like the bidcoinhistory is about two hundred andfifty gicabytes. The therm Sistory is alsoabout two hundred and fifty gagabites, but full nodes don't actually need tokeep the sister around it's a kind of common misconception among severalpeople in the Bon space that say, you know I theory and full notice toterrobites right or that therium, no that's two hundred and fifty gagabitesis actually last secure than a bit coin. Full! No! That's also Tuin TofiftoKikobye, a Goin on that quite a bit Gogood ten in I cas I don't have to reait reay too much right, but so to cut to the Chase Barn Barn. All that allthat stuff is that N Bitcoin, the UTEX alset. I think it's three and a halfgigabotes pretty maneable by Everyde Computers Right H. No, my computer at home has like thirty,two gigabyesoeratmigt but anely it's a nos stop. No Plan of people might havelow POLISS wit Aha goodbites around, and you can keep this entire state inRa, which is great because you need to do random acasses into the state. Wenever need to do random accesses in the history and, in fact, like the evmforinstance or the betcoin consensus rules, never look into the history. They don'tcare about haven in past transactions. They only care about what is thecurrent state of my viritual machine in otherium. The uncomparase stancize isforty five gigabytes, so I don't know if Marqall was onever or if you gotdiscussed Eric while thinking ov a punl note. It took Hem like twenty five days,O th Sincin Atheriand, full nowfan scratch, because the state izes offorty five Gigoodbye t you couldn't keep in Arom, so constantly ou had tobe like smopped out from his SS dsssentually H, and this is kind of theproblem is that state cannot be proved. It must be Keptin as entirety and mustbe kept in as entirety in a way that you can do fast. Roundom, acasses, H, it'sit's, it's a reasonable qualifierI'd say: that's t's, that's good yeah, and this is where roll up scills. Thisis where I get scale ability benefit from, because the state all the stateaccesses are not done on chain. They're done. Offchan. All the statetransitions are Don offchain. The only thing that its posted on chain is theactual data, and it turns out that the amount of data that full nose can justdownload hash once and then save the disk. Even on a hard desk. You don'teven Messas Dy to do this even on a hard desk. The amount of data that youknow even average consumers maptops can process, and you know they, and then they can start a disk and soonwith an average Internet connection is congut you to VSA levels worth oftransactions and itll makes all TNOS actuallycheaper to run, because now you won't need as fast as as d. You can even dothis on a hard disk, and this is like a really like paradime shift completely too what allthe previous scalability proposa that have been onfaously. Who needs to careso like say, for instance, there's a specific. You know: Cool Horde ofpeople that enter into a optimistic rol up right start using itt to to transat among t each other. Does anyone else other than that coupof people actually is at the entire set of those group of people who need towatch all the transactors to make sure everything's going appropriate and who needs to download all thesetransactions and maintain them and keep track of them because of what's beingstored? Is the state rout in the BACHINSECT? That's what you're doing inany level too UM technology is youre rooting, your trust and security modelinto the main chain below it right and so like that's just the trust ofeverything that happened is valid and we can prove it and we're putting ithere. So e can't be tampered with, but...

...in the process of doing all of thosethings, you still need to keep track of all the STEF that gets passed aroundthat there's no fuckery involved right. Who has to do that for TMISS Grollayeah, so I'm going to get unde? Who has the incentive to do this, and who doesthis h a bit later, but first wait. Your Sitwpoints the first one was who who has encenter to do this, and the secondpoint you made was Oh have to do this, like O, has Ta Ame same situation righte okay, so I should get to that later as its one of the interesting benefitsof ofthemistigal obs over, for example, Pazma variance so we'll get to a lat abit later. Do you mind if I kind of go over now? Oftemestical OBS works asopposed to dcollups, and then we can go over the nice properties, and thatincludes answer juvotion rock and roll Pervec, so aptimistic rlops, it'sactually very similar to zicarrobs. So, let's start by, supposing we have aleader and Welw after I'm finish the description of the system, I'll discusshow you can do leader, selection, at's, so supposed we have a leater. Theleader must well. Teyor could do nothing, but if they decide to dosomething they must extend the tip of the sideshine and the on on chain. What the contractsaves on Chan is essentially all the blockheaters off Thir Sidechin, butthis ror up than H, not not just the La technically. You could do it with justthe latest blockhatter, but really you'd want for convenience Y at's.Let's just say that you have all the ball hetters, so you can think of thiskind of like running a light, client BFOR, the side chain and you're runningthis in a contract, so the leader well Simita Trnsaction,and this commits to kind of the entire side gem block. This includes the ballheader and all the transactions and all the witness Tata, O all the signatures,not that anzice roll APS. I said transactions are there. This isactually not entirely true and Zca rots. You don't ee the transactions, you justneed the state transitions, so you don't need the actual transactionitself and you don't need the signatures. You just need actual estatetransitions and the ZECA AL UP, H, wh n, when generates the proof, itoll checkthat there exists our signature, so you don't actually have to provide this ontime. So you do get some benefit in terms of okay y. You can call this acompression because you don't have the post to Sinn Tha Sou on chain. Thisisn't actually meaningful because, for instance, you can use a BS, I aggrecatesignature. That is like consensized, for you know as many transactions asyou want and therefore you know the difference. Beeen, a single consensizesignature for like say a thousand transactions, an a thousandtrunsactions with no signature. The difference between these two isbasically nothing. So again. The reason roa up, scales, lots of people say isbecause you know F, comprassion or you know, validating transactions.SISINCTLY is not it's because it doesn't do state accesses 'cause youjust pols Stuch, anactions, Oha good, so in Optimiston Rele you do need toactually post the entire transactions, not just the stat transition, soyouneed topost entire transactions and the witners state. I Sol the signatures H and you also need to include a bond.You don't need to do this with Zca wolps, because I mean I guess you couldpost a bond but then Sinceega rops. You know you immediately in the same block,berify the Zka proof, and then you know if it's invalid, you just don't extandthe chain here. Yo do things automitically, so you need a bond andthen lye construction by the above since the leader, Mostexten, the tip ofthe chain is four free right, Yno forking. That is how it is by construction. Ths,not contractinforces this. If you try to extend something, that's not the tip,then thet, just dejrects your block Oa. So if a blog is invalid, anyone notjust later anyone can submit a fraud proof that will roll back the tip tothe immediately previous block to the frauduand one and they will claim half the bonds of all the orphan blocksand the other half is bburned. Note that this half is technically assystemtparameter, but it does need to be greater than zero like you do need toburn. I get not Ha gread in thes, Jo, but you need to burn a portion of thebonds and you need to reward a portion of the bonds to the fraud prover. Ifyou don't burn anny of the bonds, then someone can asuntfront run their ownfraud and recoop their entire bond, and this was something when plasm grewup, initially released their their iteration or theire. versionalauthomitic Ralups, then they had this issue an in my version of optonicion RP. Sowhen I created autoisicl ops, I think it was back in June they released theiras in August, so they probably should have based their design off of mind, but I guess it: Din't 'cause. In mydesign, you'd burn half in your loartasf especically took aoy the frontrunning issue so h. anyone can mie fotproof and, likeall systems of thro proof, there's his inkcer e assumption. You Know Belet'ssay two weeks. Ite could be anything, but you know it's a long period of timeand then after that, sidtine blocks that aren't fraudulent.We consider them valid. We consider...

...them finalized SOTF, Pallsme Fund S, hease just teck,with the Saj, an you Osun Fun, so Contracti asmented on the SADCHINE and withdrawing funds. Now this iswhere gets really easy, which is withdrawing funds you just starte withit all on the sidecine by burning some funds and then once that Siejane blockbecomes finalized, like once, it's assumed a Balad by the contract. Youcan just do the you, can complete the withdrawal non interactively on Chan,and you can do this because otegrup is permissionless and plasma's permission.So in PA MY GOS SP start the withdrawals on chain. You cant startthem on the Sidechain, because Poy plasmanis permission, so those kind ofone of the nice benefits offce, there's a specific call within the sidechairthat says withdrawl that Burns thus finess and then, when that cat finalize,that allows you to withdraw them from the spart contract of that. If you arethe owner of the same of the the person who called with draw function, Yep,okay, I'm still having to Bistol Hav IV bit of an issue with the like cnsensus of this right, l,theyaskhow things getagraight now and who get atsay. Well. I guess I thatright now Hich is like I said after I, N description we'll go to lead aselection and that's that's whet 're getting to right now so leatereslection right 'cause before I said: Okay, some leader extends he chain, butwho the hells h leader it. So the way this works is what I callmerrge consensus, and this is kind of the topic of my research over the pastyear. When said, I was doing a latter to scale buildy research, Ismergedansensis IDA and it's a consensus protocal, that's entirely verifiable onchant and kind of the way you can think about. This is that consensive protocales usually have afew properties. They have fortise role a so you know multil forks. How do youdecide to which one is in fact Ocanonical Tin, you have a block withOto Doufunction, so this is o just your state transition. You have leaderselection and you have a SIPLO resistance. Mecanism, you know PrWorkstak or whatever, and on, and you kind of combine all of these, and thenyou got security, which is the cost of manipulating the blockhaing history.Iyou know doing reors and censoring and soon right and what you want is a consensus,protocall, bith, thus properties and so on. That's fully verifiable on Chan andwe note that Nakamodol consensus doesn't work because it kind of tightlycouples the for tes rule to Eto selection and the simple resistanceright, everything's kind of tied together yeah, whoever wins the blockkind of gets to extend the chain and so on right, but the trick ers, there's nolike time stabbing it t the longest chain eventually will win out, but it'snot guaranteed to win out immediately because there's nolike time, stampinghere HIV, you have probabistic finality, which is exactly, and that is not verto the cash rate of those bloks exactly and that Os not veryfoanilasic by hatjust but like. If, if, if you give me like two blocks, you actually know onchain which one came first in Achomoto consensus, ancwer is no, they can makeup their own time. Stams right, that's the problem. SONOCOMODIC can senseis doesn't work,so we can't use just prove of work for this right, but what some things thatdo work, so the Nice thing to notice is by construction. AUTOMISTIC ROB is forfree right, so you don't Ete a fortise, or rather you just need a triviolfortice rea, which is just take the only four that exists, easy the block,the doufunction. So when Yogoff the block that I is the ZCP proof, anOPTIMISTICA OBSB BLAC validity, is you don't validatit on Chan? You say youknow, I assume it's valid and las someone Smith a fraudproof on them, acertain timeup period soy two weeks, so we don't need a bloss ot thafunction onchin. We do need to have some some rules in there to kind of verify froudproofs, but you don't actually do w the validation of every plock. So reallywhat were left with is on chain. We really needed to do a leader, selectionand SIPL resistance. So there's actually a number of ways. You can dothis and they all work. They all have like the same theoretical guarantees.So the Wy I originally proposed is that you can just do a first comforster. Soliterally, anyone who Mitsa transaction that extends the chain ges to extendthe train and that's it Um civil resustane, is provided bytransaction feese by the Manchains bloc capacity and theres works, because themain chain provides security. You don't need to have like an amazing Sibiloisin,the meghanism you don't need to have. You know perfect. Leater selection, youdon't need like now great for choice, rule that accounts for all these. Youknow sinconthy assumptions and you know and honest majority assumption and alltha stuff. There's no N, as majority assumption, there's. No one notirizingblock, there's no long as chain. You just need a leader to extend the Chanif it's invalid someone soms a fraudproof thise. Is the amazing things that th re'? No, there's no honest, majorit suption here, just you're justdoing leader selection and therefore you can do something really simple andwhat I propose is just first come forserf I feel like a you can also doeasily docible if the mainten is not thematan ascensorsip resistant. Ten, it's not easy to Dase that R Lik, you can say.Oh, I pay a higher transaction feey, but a higher transaction p kind of gets you into an eartier block,maybe but doesn't guarantee you're placed ahead of all ther people in thesame block right. You would actually have to bribe miners to reorder thetransactions and no that from like a...

...crypto economic security perspective, forcing a reordering of a transactionacross all blocksis equivalent to censoring it. There's no differencebetween the two. So I the SOY IU said Yeah. Okay got, which is like reallybizarre property is like you know. You can th think of censoring like orderinga transaction always just past the end of the ball capacity. So there'sthere's motually no difference between the two H. So as long as I mantain ecensorssip possession, which again this is our Y, know starting assumption,then this works. No, that this is just a profosal. That's not necessarily thebest way to do it and the Nice thing is it's Emergi consensus. It's a Consensi,Poto call, that's fully verified blon chain, so you can actually do anything.One thing you're can to do is randoms PR. Sick people can just take coinslike ether or whatever, and then they can ranow to generate a random numberto stelf, wl them around and then choose the next leader, and you can dothis on Cha entirely. So there's like a million different ways. You can thinkof doing this right. You can use a DF if VDF ever get maed. Who knows H, youknow you can use diferent ardom number generator or there's like a millionthings. You on Ta do ith just as long with this verifiable on chain and that's basically it that's like. Ohand I guess now we have to answer a question which is who has the incentover? Who Will you know check who who will generate this Fod prose? So theNice thing I sa, I like geting the key in invasion here, that's actually whereI am right now, so I want to give you a chance to go through the whole, likepresentation, Piel of like how of the data, but I'm not reallyfollowing with the key innovation years with the roll ops. So maybe you canhelp me understand this. All I hear so MMSOWELL. I got look up for that, yeahthat yeah so like I ' so there's like all this talk of whatwe're committing and who's you now, who does leader selection, blah, Blah BlahBlah Blah okay, but that's coololl. Where do these transactions reside yeah? So we'll got to that very shortly.I guess I'll answer coraus point first, which is that you know who kind of hasthe incentive or who, who will h. You know check for check for Faduleintransactions, O foin. He blocks right, yeah sin NS alsopart of my question waswhat Collon ask too as like: Where do these transactions to live becausethey're not getting put on the block Chai and who asand, who has Tel, holdthem and manage them? What is the role up mechanism? Where isthat roll up being put? How is that roll up being maintained, N- and Ithink you've gone over some of that and I just it just didn't absorb for somereason. So maybe you could help melakeup I'll cover that after answering the COSquestion Birst, yes I'll cover that so to answer. Core's question is thatsince the since the permissionless, right and since is fork free, if you'rea validator and you build or like a leadr or whatever, and you build uponan invalidibe, lock, eventually someone's going to summit a froud proofand take her bond. So you don't want to build upon an invalid block just likewhen you're in mine and prove work. You don't want to build upon it, involveblock 'cause, then you're, Balkand, macket orphan right. So every single block producor in thesystem, is incentivized to Vala toblocks their building upon, just asthey are nochamotican sensus, that's kind of where then soundolize. Butanyone if they wanted to be absolutely certain of you know the security otheir funds could do all the validation themselves. Clienside H, which bringsus to kind of one interesting point, which is that H, one common criticism ormisunderstanding of the optermis to grow ups is that it introduces highlatency. They give it away two weeks before Yeu transaction pinalized andthat's not entirely true, because you can actually do clientside validation,sincefour Tree Youre Guarantee thetha blocks will eventually finalize andtherefore, if you do clentside validation, you can immediately acceptValuabe ocks as final, because you know that will eventually Finlie so like assoon as ou see a block it ies the same security guarantees as a theory, I mlike the same cost of reor, so you don't as you to Wai two weektoeptetransaction. You can accept it immediately and if you want t inserin with thedawls, because you don't want to wait two weeks to withdraw your coins, youcan actually just use atomic slops with hequidiy providers. So someone sayinthe main chain can provide liquidity and will atomic swap you out of theSidechan. So they escentially now have funds on the Syjane uffuns on the maincrain. They n t the charge little fee for those equidity and they don't mindwaiting to weeks to withdraw a bunch of the money h. So this kind of opens upan ingesting market for the quuade providers. They can provide theiservice trustulty and s completely trust us, because they're, the Kitiproviders, no their atomic swop, won't kind of getreorged under them, which is a big problem for cross chain. Attomi slopsright. What happens if one of the two chains kind of gets reared and thenhalf of the Tomin slop gets just gets indalidated with OPTITANE ORUPS? Youdon't have this issue. Tho blocks will...

...eventually finalie so with clnscievalidation. The lequitting Priter Knos like as soon as you commit to your halfof the time, O slop. They can come into theirs and complete. It is there. Maybe I'm missing this part. Are wetrying to have an auditable side chain where I can always see the history ofevery single coin and when it was when it was put on chain n when it was left?Yes, okay, where do all of those transactions live because tyey're notbeing put on somewhere yeah? So all the transactions, the transaction datalives on the Mainchan that doesn't increase state? It onlyincreases the I guess. If you would like to run a full mode, it increaseswhat you have to have in terms of well, I'm trying to think do't ash increase.What you have to have, because you as say a main chainful, no do not need tokeep history around thos kind of one of the unspoken evils: a Blocka, it's moreof like Iol, N Tho right those Lik for th for the blockblock explorers thatwant to keep the history of the ethic the entire history of the Blacchain, sothey can perform lookups appropriately m teyld need to t carry tothers in ofthis Ye, so I mean the reason this is also not an issue for Bakisploer is. Isthat so I guess so Arvo, I guess Ar Hav. No,so this is you know wher, I gots confused Ng, because Bi coin, in atheory, I mean use different NOMAC ature right, so it to be clear. Let'ssay an Arcrab Nowood for atherium right. This is the one wor after every block,or I guess maybe after Avi Transaction. There is a kind of a snapshot of theState of the system right. The problem of this is, it means nowyou have this enormous Napshot, that's like way bigger than a block currentlylike this is forty five Gogoodbye it's right and you need to keep the snapshotaround. That's why block O now Arcanois in etherium are two terrobites in size.Right 'cause. You need to keep like th ou know huge snapshot of the State Very often so imagine, instead of having twoterrobites, which is just basically wasted. 'cause tit's, like a lot of herDunan state elements. Imagine if you had two terrobites up just a rawtransaction data that'd be much easier for Blockin slores to handle right itjust just history. I just you cal put it on your disk and it's easy Pz rightand there's no, like you know, tight intertwining, you don't have to do abunch of like weird Dandom acasses, just transactions od like or be able to search thatefficiently. You'll still need to index that intormation appropriately rightand that's for the Jority of the weight of Arceble notes as as the index is built, so that youcan query it approprately, because the way we still want block ten Datas isnot built for quering, so, like Ha, you would like to if you would like toaudit this type of a little too future. Something is going I to be built sothat it coan be queriable. Now that that information will exist on theblock chain dus, not necessarily in current state, and so like that's her,it's where I maybe see patetial hangups. If this its big is it's gin kind of like what it is for atheory. Im Right now may be a little more exacervated. It's going to be hardto find historical data for any regular user. I would say so, in my opinion, at leastI mean different people, ave different opinions, but in my opinion this isn'tthe problem because we can get or like if all you're doingis just you know, storing history in UH. They just putting on this on your harddesk h. Since you don't have kn w all thesr state accesses this, not actuallythat expensive and even if he innexed like blocker, for it's not thatsignificant yeah, I'm not worried about that. So much right. Now, I'm stilltrying to work out the fundamentals. Here we didn't realize I prhaored verydifferently: Yeh O what is getting commited to the chain yeah, so on Chin,what is going to commit it Itas all the Sidechan transaction data, alltinshos,actually data raw transaction itcorrect Li gues is th. The side chains couldhave. You know thousands of transactions that you know. This is aroll up of that side. Chain data correct, so ith, optimis IRA OS is theraw transaction dita, and their entire to on on chain is being yep. Why? Why isn't that? Why? Why is that better than just usingthe Jin? Oh because this is cheaper so, like I said it took Eric Wall to sinkcase, a theoryum, a pull note. The thing is tha theore and fall. Not Ittook Eric Wall like twenty eighty five days right- and this is like twohundred fiftiaobites of history- the state size is forty. Five gigabytes, if I told you if all you had to do wassay like download one terrabite of data and just download it and nothing else,and how long would that take you approximately you think depen Om, ICana downloading and noth, not evey Dodo State Transista on all of that, asiswher. The savings comes from yes, I'm Worri,like not as a cost comes from. The storage is quite expensive on theblockchack, only stays torage a history storage, so so so here is the kind ofthe evil like I was tsaying. Is that block changs Rean algoism? You canactually cat price history, storage and...

...protocol because you can't enforce thatpeople keep history in protocol. I it's like Ife, you charge like whateveramount of Gasan say people are going to keep it, but people don't full nodes,and this is the evil that people don't talk about: Wh, N and even Bitcoin Lis,an eltras and thi full nodes. Don't need to keep history, they can prone it.They only need the latest tate and maybe, like you know some previous.They can keep it another BAC. I Pere Right. So so, like okay, I'm still notgetting the very basic like current state process. Okay, I have. I walk methrough a user journey here, man I am, I am going to join a site chain,it is Adlerchin, it is your Testipang and I I'm gonna, withdraw of some East andput it into your sidechape. Okay, Corey and every member of statusdecided to also join this, and every member of cassensus also dosides tojoin this and we're all going to like swap our a theam around for coffee.Whatever we do this every day about five hundred transactions go through anhour: okay, no yeah, it's a reasonable amount. It's not ridiculous! L May TATwhatever M. When are those transactions committedto chain on demand immediately, it depends on how you do leader,selection, okay, Sav, the the first person to publish, gets a block yeah.So in the first come forst serves an IT's. Basically, whoever wants to itcould be within this participant Seid. There could be, you know outside theBriti spenset and there could be aggregating in transactions and somesort of peered to Peermampool yeah. It could be, it could also be not beerdopeer. It could be like each aggregator just wants to like have their ownserver their own, like local pool of transactions, and you just like submitsmit thatransaction to v like some tp reqlest or something and like yheiaction right, there'se different ways, but you know they have the Trinsttactall y aggravate they commin on chanp yeah, no someboy. Now somebody decidesthey just get to published transaction and then like theyre tra. Somebody elseis also trying to publish at t the same time they have half of each other'strandsactions in there. The other half is something they maybe didn't decideto include for whatever reason, what is going to separate those two and how dowe like wha? What would make that valid? What's doing the actual checking tomake sure that you're not I mean, obviously, that would be a double spedfor half of those transactions, whats, actually verifying that that didn'thappen, the front proof the fraud proof on chain, someone sommits a fraudproofthat attask to or that you know, showis that Rp not Cho, aapproves, thatsomething that violated that consensisse rules was committed to OK,now that they're submitting it to the chain Andeform of a transaction. Sowhoever decides to pick their transaction and put it onto the a blockright there. They don't know that they're running a sidechaine. So what is what is actually the do so?aggregators I mean they do know they have to run like a sideinful note, sothere wouldt know there and Igo arium on the etherium side. So the point Nosothat the theorm theres there's a beauty is that this is outside cansenses ofAtherium, like the teor in Consensopoticol, is not affected bythis in anyway. This runs entirely on Chin, so the her on minors, don't carethey're, just tatransactions see this is this is the confusing part. This isthere's two chades. So when, when we said where how does it get committed onChan and all this stuff, I I was talking about a theorym Ye. Do, Oh youwere yes, so what I st S, I I would not I wat. I was talking about the sidecineOkalet me rephrase, then, okay, an we're going to need to make thislanguage a little clear idea, Ye j, don't say Jane. I think we're both verysmart people. I just think it's it's verit's like there's a lot of data hereto injust and I need to get this. I need to get this language down, sowe're going to talk whenever I say Chan, I'm going to say etheorium and whenevera Sicen in just say Sichin. So transactions are posted to the sidechain. They do the roll up there or do they do the roll up on etherium, sothey aggregate their transactions into blocks on the side chain, okay and thenonce they have once they've constructed a block, they committed, Oitheria, okayand so who's posting it to a theory, the leader of whatever, whoever decidessupposed to correct s, ill get cool. So somebody just randomly says I got ablock. I like this block, I'm going to post this, for whatever reason right, Iwant to handle this now m somesomebody put out with draw from a theory request.Therefore, somebody's got Ta Hande us, I'm just going to go ahead and do itright now. Rightnow two people do that in the same fourteen second time Stanand they reside on the same block. Well. This is why the first come for Servisnot ideal, because in the first come for Survyeu of race conditions, wheretwo people try to extend from the same tip. So the first one who gets in isthe one who extends the tip. The sagond one instead of block is rejected right,but the first thing Oesqarehas, the availability has an availability bonus,meaning that if anybody could post the block, then you don't have to wait onthe leader or make sure the leaders constantly up somep right. So...

...what about using a classical consensusprotocol to do Leder election and if that leader becomes unavailable, youcould stop the leaders that way. I all ounching is this all Offchan. Theyrun raft when the leaders elected they go and use that as Aler at go right asas dlike. You would like that part to live on shame, so the leaders electionneeds to be do on Chinese to be verifiable on Chan cause otherways. YouR line on honest majority assumption that reports to some correct executionof the later selection- and you don't want that. You don't wantto reon an onest mjority sumption of S, Apaila bl, a problem with the leader.Well, oh even available a problem like if the leader just goes offline. Well,then they lose money, they loe some Stak right like say you do. RANDAPPISPROV stake right. Her leader is not online and the you know they're assignas the leader for the slot at the period of time, and they don't come itto a block. Then they just lose their stick. Yeah, that's fine, but that's not finefor the people that are trying to like. So where does? Is the money everbrought off chain? So is there any sort of like committing my money to addler chain so that it islocked up and so that it's in your smart contract and how do I exit incase you go offline because you're inernet out 'cause of the weather, likeyou got hit by a hurricane, you couldn't keep your shop, so you justmake sure that the UH do you make sure that, as long as theleader selection is permissionless so that anyone can join yeah, you justbecome a leader and you create your own block eventually. Now the interestingthing about these protocolls? Is they basically allrely on this thing? where,if you manage to do something and find our time, then you're good, but th,don't care what that finatime is. It could be really long, AIS 'nfortunate,but when you do these, you know proofs of of the security properties. Isthat's what that's what you kind of have h disconnect for ly? What most people Idon't nk at least my understanding of level two technology is that Um, it's alm, I feel as though it'sfundamental and that it takes time to get on a chain, and it takes time toget off a chain once you're on it. You cand do whatever the Hell Yo want. The main difficulty of all of thedifferent solutions between level Tu technologies is how they held. Do youget off a chain in in the most trustless and available manner? This iswhat John feels to be the best option so far, yeah so to to answer that, interms of like depositing Topositin Y, it's usually easy just to posit depositinto contract. This is basically easy for all f the proposals, as withdrawingthat you know takes various time. So the Nice thing about it beingpermissionless is that you can always. I guess you can rely usually on theexistence of aquodi provider that will atomic lop your funds out. Colly providers there you essentially,as from a user's perspective, get your funds out immediately. You no longerhave to wait like along withdrawal period, and this is very powerful. Wtell me more about this lequidityprovider. What is what is? What do you mean by that? You just have someone that has a bunchof money, and so this is actually where things get really interesting inchannels. Let's a you can't close your channel and like give give your moneyGiv you, the funds in the channel to someone else trusts they and they knowthat they're going to get at trust, aty and all that stuff right. It's like youhave to wait until your hannels close and then your funds are unlocked andthen you can do stuff o you can't like. Do you can't do an atomic swolp with ea Kodoo provider, because again with the normal channel, if you trying toassign ownership, are you trying to like assign someoneelse to be to represent? You is kind of one of the participants in the channel.Then you can double spend this. I do t Li you non PS, Mycashoy in theblockchand to order these, so you now the normal Chileld. You can't do thiskind of atomic slot, so how d how HOS kind of work? So someone has a bunch ofmoney right and OO, someone? U E A bunch of money, they're on chain rightand they watch and they wait and see through some communication protocol,that someone wants to witherdraw funds from the authemistic Ralps back to theTheram Chin right and then they offer for some for some Cott of the moneyright for some percentage. They say: Hey, IWILL atomic, some OP, your fundsout from that side Jam so that you know how funds on the main Chan, Fon, thetherum Chan and I'm going to have yourphuns on the on the Sidechan and the Cuto vider's. Okay with US'cause. They don't mind waiting two weeks to withdraw their funds. Theydon't care, there's just a a bunch of money sitting there right and this isrisk free. They can do a clientside of validation and they know that a Valibloc will eventually finalize. So this is risk cree for them. It is literallyrisk free money, theyre paying Otaneer in mine, Tais Te withdrawl, they'll use, onl Owoul, take youfrom asecurity standpoint, two weeks to say I have my money back on the man Chan. Ifthat that's, if that's the time period, what people are doing is they have bothmoney on the main chain and money on a sideshade, and they say all right. Giveme your money on the side chain and I will give you money on the mainchaininstantly and I'll just take that and...

...go through the towk process myself forthe fee. Well, they don't even have to do the two gir Hankan Jus leve up. Itdepends on whatever the whatever theyre they're, knowing in terms of being aprovider Ke they'r on hurry right. They have a bunch of money just sittingthere and they can use this money and make money out of it by B sentracting.I the service perletter and is whis free, certainly interesting, and I I feel asthough I I I need. I want to and need to know more, but s, actions that that are happeningon the chain. What is the nature of them? Are they? Are they exactly Um etherium transactions on teside?Jenuman? Yes, Oh got yeah. I did that against my bed.Are they exactly a theory in Transactsitol on the sae? So this iswhere things goet interesting is what kind of fraud proof do you use andI've been talking to the Arbicram guys who are working on this interactivevarification game, roll type system and there's kind of three there's threetipes of roll APS? According to their classification, which I agree with,there's non interactive or all ups, which are common, tly called zcaroops,so theens here, there's no interaction: Ig, you post, oedge, proof, youpols thetransactions, you Veryfi n one block and then Er's either balader involved,and that's that right then there's a single round interactive roll up andthis is called oftenist to grow ups. The reason hi Singaaround is that someone first commits to a block, andthen you have a single round fraud proof to andalidate the balk later, butthere is theyll techically interaction. There is a San onassumption when you'eusing frod proofs. So it's not completely non interactive. The fraudproof itself is not interactive, but the entire system is not notinteractive. I has a single round interaction and kind of a lot of when I createdAUTHOMOSTICA OBSA lot of the designor work I did was to make it noninteractive or to make it single round interactive, rather because Pasma cashhas multiround interactivity, which is more vulnerable to chain congestionunder certain anditions, in addition to being horrible, Ux Etctera. So the third classification number L,what the Arbotrun guys are working on, which is very interesting, is themultiround interactive roll up, and this one is where o using a trubit likeInteractive Verification Game Right, where you kind of, are you ay Ramiliar,with youbut yea? We Fed them, morfedacs, no prefect yeah, so so use useinteractivarication game and it his benefits, because you know you can doway more complex calculations, an this intractivvarification game than you cando with fraud proofs. So I regards to what the transactionlook like. The answer is, it depends on what people do so one example is fuellabs, which is pretty close to launching their a public test. Nut isusing like a utxo day to model and very efficient and compact froudproofs forutx Os uh, and then so you can do things likepayments. You can do exchanges, you can do stay th, predit, cotd script, Ting.I E KINDF H, t kind of like what that coin allows. You can all anfeol kind ofoffers all this, but you can't do UNICLOP on thisbecause you don't have like you know a nice account day to model and all thestuff right. So if you want to build Uni Slawell, you have to use some sortof you know. GENENAL purpose execution engine of some sote right and kind of the naive way that Yourethink as well. Let just use fraud proofs just like you would unless usethe vm on the sidechine so that you can, you know, use sledity and use all thesenice tools a know. Everything witl be great. It turns out that's a terribleidea, because you can't easily verify evm inside the BM, not interactivelyBuke. You can't do a fraudproof of evm inside the M. It's way too expensive.You have to make a whole bunch of changes to the point that it's nolonger even the same solidity is no longeanything. I don't know, there's any groups crazyout there that are trying to do this, hopefully not h, but the otheralternative and biable approach. If you want to do general purpose, commutationis used. You know the multiround interactive roll up. What Arbershom isdoing this allows you to basially do anyarbitrary computation. There are certain. You know things that there's sertain trade offsbetween using you know the single round interactive row ups, which is the frodproofs and versus using the multiround interactive rllaps, which is you ow,the INTERACTIV, arficacation interacte verification game. There's certaintrade off there in terms of like how scure are they against chaimecongestion, lit the Uy around these and so on? But if you want to generalpurpose computation, then you can do it with. You know this interactivearification game. What the Arbi Tom guys are doing and I'll probably getyou fairly close to you know running therumstyle, smart contracts, there too, so tans, I guess kind of Clr F. This isthat it depends because the Nieto mister grow ups is that as long as youcan generate either fraud proof or an interactave varification game. Now youcan just do whatever you want. There's no restruction on the particular trenstation DPARTMENT, when Trad's eens are posted to a theory,and what does that look like whenransaction ee postad to what the AMh? What do you mean by? What does that...

Lok Lick? Is it a like? What is thenature of the transaction? Is it Chane? Actually containing data is n Po do nowcall that as yeah yeah yeah, so it's basically in the Caldeta, you just have all ju. Have the Saan Book Ave aShitloer Caldata? Yes, yes, and this is a new design, paradime USIN Blocchans s.The data vabilty Laer is kind of a new design paradim that people haven'tIsaren't used to and there's new projects, for example, agy ledger thatare build from the ground up to offer massive data availability throop, butwe're talking like tens, tens of thousands to hundred a thousand oftrums axo o second, all wit, aretaing the same security guarantees justbecause they fobus ondate availability. But this kind of scheme, where you knowyou use a transaction and use a Bouchan just for Datea availability offersestainable scaling and it's to a much greater extent without increasing thecost of full notes. DIS disproportionately and to that end,I've actually started writing up some IPS and a Britan. Some proposals, oneatresearch, to improve the way a therium works with you know using the chain isthe Dat avaiablylaer. THE FIRST ONE IS EIP: Two two four two which introducesa new field in the transaction, which I call Postata NN subject to Charnge, I'mnot great at naming things, but at the sentuly four data, that'sposted on Chin, for these kind of you know all types games and the poste IDAis not accessible to DHEBM. CALDATA is rigt. The EV m cnoshe read from theCALTATO and loaded INOTHE BM post data is not so you when you have atransaction on h, postata kind of the client w will know. There's thisinvariant here, there's a promise that this Postata dvm will never see it. Soyou never have the process and all you have to do is just dump at the disk andyou'l go to go and intentially perinted. If you don't care about it, yes, aslong as well, if you sign a nice o so that instead of holding the wholepotato, you just have like a hash of it and near Tun theaction. Then I guessdefinitely you could just prune it and fo nore could just keep a hash of theTATA and then still be able to like valad it Causei'trtryg to Pickut.Imagine a world where this is. This is the scaling slotion right, everyoneuses it and we have a tremendous amount of cohorte of people rooting themselvesinto the mainchain of theory. Em. This way right and just jumping a tremendousamount of transaction data into history, basically, and then who's going to burythe burden of carrying all of this and keeping track of it, and in my opinon it can only work. If notescan selectively choose to keep or not keep this information based on whetheror not they care about it. EAH and Beton to Constructe e solv or anmor eTain te exports at so talking over each other N. now. Can it calling Goahea youneed come of a subsubnetwork model. I mean n an I'm, definitely biasdtowardsthat for obvious reasons, but it's a it's. It's just Um. It seems to me, like I, don't want tostore your crap and like tha, like Hee ta Te Light of Alitharium like it's orall wachane right now, because I I'm like I'm kindof like do I actually careabout this sidechain over in m. You know: Kenya, Here's, the DA here's, thedifference in Orl up an what current block chain solutions have offered sofar is that I ill get storing it. I don't want tohave to process it and that's, I think, that's for that's.Where roll up it gets it get s its Otoce scalabilityfrom. Is You no longer how to process all of this transactional data? Everit's just stored on the main chain yeah and I was kind of getting to this earthere, but then I guess I got to complete Ta Sidtrackd, which Iapologize for, which is yo now just downloading, even like th, twoterrobites of data and just putting t on your hard desk is not expensive andlike even block as porers, won't find just doing this hard or no user fullnotes like unuser ful nodes, there's not an expensive thing thik. I was justlooking at blackfried a deals, and I saw like a ou now twelve terribi harddrive, and it was like four hundred dollars os like absurd amount of whatwetinue toget better and better an better overtime. Exactly her a Donsta and bandwith McDonald band wih is just going to keep getting better and better f course.I say she ut not so much for you and Ram, especially 'cause again like youneed fast, random, axesses an to into the state. You know, ramspeeds and Ramcapacities have not kept up with things like you know, parallel processing capacity and thehard dribe harddes capacities and bend with. So what is the idea like ustcase wasthe ideal scenari like usage senrio? What is the through put you're lookingto get through these side channels? Soon I change a col a so when ful low launches I'll probablyhave a two undred tons action for a second on current atherium. Okay, whichis you know, substantial improvement and again wi'll. Do this sustainably?Is that nowite no longer has to worry as much about token state, bloth stayeblow from tolken transfers and then with some minor optinizations, againWego with the POSTDATA and so on, and you can add some extra stuff on top ofthis post data, which I kno did some research post about. You can probablyget the number up to about two thousand...

...on axand per a second on etherium againwith the same nice, trusles and knodesentralize permissionlessguarantees same security and all that and you can get about two thousand trenaction per a second hisis. Just for a simple payments, though right in termsof Oe r computation. That's that's a different story, but you can stillprobably get a few hundred ten action per a second Sicichan state is still stored. Let'syou say it's still stored on like on chain quote unquote: Youre onthe Sidethan, the SAG, an full nos will have to maintain state for the Sidechan,okay, okay, okay, I I I was under the opression that you wre, saying that youhad to mit all the transactons to the to the therium change which OndactianGe to ate or decan is mintained. So, like the state is maintained on theside chaid. But it's the data associated with the transaction wherethey live is, is stayd on ther t maintain okay, yes trasation. The states lar only happenson te sideshade in tars of how things are processed and maintained and keptup to date s you understand what the current state of that Sidechang is thathappens on the sidechain and whatever nodes are running at Sutchang. The dataassociated with what built that state is always going to be on the ETHEREUMMainchain Yep. That's a lot of data! That's what I that's. Definitely what Iausing oth dog chains in this paradime is a lot of data. It's a normous amoundof data, but it turns out this dousing bloll chains in this way is sustainableand it's actually highly scalable on Atheory, though so, because the theoryis making the necessary shanthateip correct is so that you could actuallydo this, so that would write to file instead o right side on the side,instead of like actually necessarily committing it onto the chain, but thenhe so so. With this poltdata again,there's the first Tel, probably four ps you can do nice things 'cause. Now youhave an Invarianto, you have a guarantee that is not going to beprocessed by the EVM, so you can do nice things. For example, you canprocess it Tloi like mercalize it and then just mortgalize at clientside,multithreaded and parallel yeah t just allows you to assential ave likeparallel or multi threade data availability and thiswill- probably getyou hif you do. This they'll get you to two thousand in actions per a second.It helps quite a bit. Yes, we got e rap tyself, we're alreadyalready epased an hour um by no means of my like God at all. Thisis a solution. I definitely want to learn more and I think, there's more tolearn. Where do people go to read the research that's been put out and wherethings are going from now? Yes, so following mantoiter is probablythe best way as long as you can tolerate my trolling and crass sense ofhumor, some people do not. Alternatively, you can look me up inmedium. I've written a few posts on after segral ups or most the bast placeis probably east research. I have written about half a dozen posts sinceI created alfitealups and then later on, ipproved it that kind of outline thegeneral plan towards you know, making a therium. You know a high throughboMultiredo date, availablity layer, well, ee, I ou know ace, gets through the Pgovernen's process h and where else to follow me, I guess that's about it, butif you want to keep up with projects, so I'm involved with fuel labs which isagainthey're building. You know production grade optimistic, rial, UPS,specifically targing payments, so you can look AP, fuer labs on eithertwitter or medium and I'm Ason in Lazy Letter, which is anew blockchain. That's specicially, built for high through potatoavailability that can be used for schemes like authomistic roll ups andothers that will get us to you know tens or hundred hundreds of thousandsof transaction por seconds so Yhou're gon, an Lazy Leger also asOmbopostentte that you can look up all right takes for Texna show. Iappreciate that I definitely feel better about my mental model aroundthis and where to go to learn more, which is the goal that I wanted to Gevefrom is so thank you for that great thanks for having Me Aorian Collin. Iwas REAHT TO BE HERE.

In-Stream Audio Search

NEW

Search across all episodes within this podcast

Episodes (108)