EthCC 2: Awa Sun Yin

hey everyone um how's the sound at the back I guess you can hear me all right I'm assuming so people out the last row can you hear me great hi everyone my name is aya and today I'm here to talk about variance before sake so just to give a little background all for um for people who don't know me I am one of the cofounders of Krypton labs and before that I used to be a researcher cosmos and before that I used to be a data scientist and so for engineer edge analysis so you instantly be the perspective and coming from Krypton labs is a perfect validator a goal is to become a valid for any reasonable and public networks that would lie on prover stake by reasonable I mean that they use a very and the prefer spec that actually has something at stake and by public I mean that it is let's say arguably easy for any entity or any peer in the world to become a validating node and interesting enough when you apply these filters yeah and then first you look into like actual active Testaments and/or systems operating production there's really not that many protocols that are applied right yeah so as a company we're fully based in Switzerland and since last summer we well we started to work a lot on community development that means in learning about these protocols and then also writing content explain it to the diverse trucking community and then how they actually works so they can understand what an economic or security implications that they end which is why you will see that my oxygen your very electrolyte more like informative as opposed to more opinionated recently we move towards protocol development but still we're gonna continue try to explain to more people about different protocols so the agenda prepare for today please separate into two blocks and the first block is again informative the second block is more about me sharing some experience and me sharing some perfections around the experience I gained as a random validation services professionally so the first part we're gonna be start by revisiting some basic concepts and you'll see later on what I mean by revisiting and then then we're gonna look into some variants the first a mainly four of them out of there more and then comparing three of them first of all with the consensus the year so what choice of consciousness algorithms the tabular system on which letters important because that heavily influences on how you design the economic model behind it and then I will generally be the power like well how I view some challenges letting protocol designers will first will face in the future and then we have time we will open up for questions and answers normally the most useful part okay so I start with revisiting some basic concepts and so first of all consensus algorithm that's not equal to proof or something and this is this honestly didn't matter until recently because all the networks will item before it right most of networks and in the space of public blockchains work before work so there was no need to distinguish between these two because prefer word came also inherently with nikamoda consensus however to make your lives easier in the future because you will see that i generally suggest that you clearly differentiate between these two concepts because they actually not the same and that they cannot be used interchangeably so consensus algorithms are qualit really defined as mechanisms that enable trustless peers to agree on a specific state values and instead of values could be let's say that a structure like a layer so transactions a block it could be something else we refer something is I would say more correctly defined as those mechanisms of the determine which peers are allowed to participate in consensus so in other words which requirements do you need to satisfy in order to become a validating peer in a network so classical to custom-made major families right now of consciousness algorithms is Nekomata consensus pervasive in all proof-of-work systems and the comodo consciousness is actually very simple consensus algorithm it simply states that the chain or the the chain that has the largest pool of work or lie just pull off consumer electricity under search is a canonical one very simple now in the actor run 2014/15 start to be a new family of consistence ever greater than star to be explored I call them family are visiting fault tolerant consensus which is inspired from decades of exhaustive research in CDC British systems the classical computer science research and the different variants you will see later on but what what they have in common is that the way that Hermann consensus is as well as two-thirds of the validating set agree on that value then penalized it's just a different way of agreeing so to fab too many families of prefer something algorithms right before work is you know in order to participate in consensus Nazi earn the right to produce the next Spock and earn the rewards and it's competed with other peers and solve the computational puzzle before who prefers improve or steak again there's like a lot of different interpretations but what they have in common is there's this basic idea of if you want to participate in consensus you have to allocate a certain amount of wealth or value as a collateral so if in any case you deviate from the protocol that amount will get slashed interesting enough with introduction of different reputations and prefer state research advances the nothing at stake problem wasn't was appeared so nothing at stake problem is a characteristic that certain variants of perverse sake systems have and the basic idea is that notes in consensus knows improv or snake they can simply generate as many Forks as they want and they can like maintain them in the in the future basically no so in before work this wasn't an issue because we had support so the specific chain you wanna build on you Hashem power so which meant that if it wants many different forks I had to scale my infrastructure horizontally right add more that's more more hardware and so on you know it's valid more it's mind more on different chains but in prefer steak creating the basics of the the work that validators validators do is simply created at the signature these are the other station right the later gets broadcasted which is really cheap to compared so an example a classical example that you give on where nothing a stick problem is present is by the way is present in and other projects as well both in academia both in like commercial projects this example I typically give which is delegated prefer stick using by protocols like yours the basic idea behind the evaluated prefer stick is that a plug producer would not get voted again if it misbehaves right so let's say I'm a block producer years through new one and then I cheat I will just simply not get voted again and hence let's arguably I have my reputation at stake but I wonder if that's a fair assumption because it really depends there's no notion of stake in delegated prefers thick that's first of all there's no slashing conditions there's no allocation of wealth unlike as a collateral and the other question is is there notion of accountability and I as a stakeholder can I prove or can I first even detect that any node is misbehaving just some food for thought let's say so I guess here I go how many of you are adapt developers actually could you raise your hand looking for me to get a sense so um okay cool thank you so I would say mmm I would say especially in the theme community generally the like something meant is that security you don't have to worry so much about the security of the network because if we'll provide it just simply build on top of it and that's fine but I would say care about your security if you're thinking into like looking into different platforms as well as for example scaling or looking for something that fits better the security requirements or like those requirements of application I would say carrots and extend where you see those like a necessary thing but of course you would carry a lot more about you first build a new application better because that's where your competitive advantage lies right an example I like to give that is outside of boxing world is that every game developer I don't play I really like pre moody player games I don't play a league or dota because I know that infrastructure is truly secure right I thought it might have most of us most resilient our infrastructure can start attacks possible I play because I think it's fun so here's adapt developer you might want to prioritize that over security but what happens if I find out one day that League or Riot Games or valve has been cheating behind and actually the ranking system was not fair then maybe at that point I will actually carry right about because that's a necessary point to make you funny or you gain fun just an example another concept born overview which might seem like depends I think a lot of people just like assume there's but another novel thing about perverse take systems you will see is that they were split between the block production mechanism and finality mechanism switch again in perfor it wasn't necessary everything was in one right but now with the new variants was time to see that you have you can have one system that produces the blocks that bounces the chain and another one to finalizes them and this is where finality caches come in so if another guy just for example Casper FFG polka-dots AFGE and such you can even so you can have a underlying chain that produces the blocks and then you can have another additional chain or even an overlay on existing chain to produce finality so again now I'm going to be gonna go to some variants I prefer to think so like when my no means there's a lot of more than theirs so I just chose them because I thought at this moment people were more interested in this but you know there's more than there's still some variants are liquid prefer steak which is used by the protocol tense for example one that prefers steak is used by cosmos and any chain that uses they still using the cosmos as decay like Irish net nominated prefer steak is the very end that picado will be proposing and Kasparov if she is the beacon chain that the theory 2.0 is working on and you can see that we have first of all the economic model in on top of it but they can couple with different consensus mechanisms actually Tasers although its purpose Nick where there's actually something at stake just like a modicum sensors for block production and also for consensus whereas wonder prefer steak is going on full tournament so they are using vfc consensus and not only prefer steak and Caspar is kind of like a mix between both you can use in the line broke production system that relies on the Kamata chrome sensors and yet have an additional gadget the price for now let's see the code will I could be a variant of perverse sake sir of visiting fault-tolerant systems so let's compare them first at the consensus layer actually left out Casper because I'm sure you can find people who can explain better than me at this point I so and I think it's important to understand a little bit the consensus choice because that actually influences a lot on the on designed economic system behind it the first of all tells us tells you use the tech emoticon sensors Nakamoto consensus is very simple like we've seen is pervasive everywhere you seem very simple in integers validating notes actually have two tasks not only in producing in validating blocks but also in generating endorsements and those rules are simply this operation of I've checked app rocket is height and I said this despotic script like a validator spoke just like an additional attestation and consider the network could be partially synchronous there's a chance that when is your turn to validate you have to choose which like which chain because there could be there could be temporary works that are generated it simply because there's some Layton's in the network and then at that point what you would do is you would choose the block that comes with the precedence to have the highest amount of endorsements she's like simulating you know it's like simulating proof of work without having to do the mining part in cosmos which I'm really excited about because I think a lot people not realize in this but sedate there's no network yet that is using Nakumatt there's using PFC consensus at today there's like well Versailles Cyrus but up to date all the networks of light on the Kamata consensus and tenormin is going and say we only finalized it so we're producing blocks using the same system and we're only finalizing them I so as soon as two-thirds of the validating network of the of the validating set are provided at the station so like if two-thirds if not enough are online we're just gonna wait and then and then eventually the hopefully it will come online Elsa we get left out from the consensus set and eventually have enough threshold to continue the chain also we're just gonna hold the chain polkadot which is working on hey so people call this grandpa or shaft and I categorically refuse to call it like that I call it I'll spoon out the gadget my preference would say so I was finale to catch it I'm pocket or there would be and the Lime Rock production system most likely relying on on a variant in the kimono chrome sensors and and then there would be the task of the there will be another task for the validators in the relay chain which is to produce finality in this case this network our synchrony or deterring relate to see you will be able to when it's your time you'll be you will have the task to vote on the block but the blog you're actually voting on the prefixes because the block comes with a set of answers right but then what happens is that you might be you might be voting for rocks that had different ancestors but then in that case the ones that only get vanilla the ones that get finalized and only the ones that actually common in 2/3 of the network so let's say in this case that it works right but the ones that only the commenting on across all of them it's only zero on one so I'm just gonna finalize those at the moment and let the change go on and then eventually there will be more common ancestors so I'm gonna get a free knife analyze those so they're like there's a gap for probabilistic for now let's see in between them I mean you might have realized that in under the assumption of perfect networking conditions or network conditions actually both tasers or both polka dot will look very so the progression of the chain we actually look very similar to tenement just an average elevation I had when I made the time in the graph so comparing economic models this is like also a very flash way and trying to be very quick and explaining how it works and I in the prefer secretaries works and I other networks so I'm gonna be skipping something Nicola T's um so in liquid perverse take or integers if you want to become a validating node or participating consensus there's a minimal requirement of you have to allocate 10000 X to Z as a collateral actually this number might change in a few weeks currently there's a launching governance process ongoing when people are still voting on it so you might somebody propose to change that Valley it's lowering so maybe we changed in a bit the way the reason why people want to become validly there some tensions of Baker's it's because they get the they get to earn the portion of the annual inflation of the network and that it's currently at 5:51 percent in the best side in the best case and the way to get to so how do they like choose who gets to produce what blog or who makes endorsements right and the way they do it is the user system there's basically so the random system there's based on how large your steak is and steak is both your cell phones so the amount I put at risk plus delegations which by the way are not at risk in the case oh yeah an operation that the liquid rivers take rewards for is in a trees and other people so let's say I at some point me I am a validator and at some point every six blocks they actually have the same height but at different and they were signed by the same validator skis I'm just gonna include an evidence into the network and broadcast it and then that note that made that the community that safety fall would get slashed and in that event me who submitted the evidence I would receive 50% of the entire amount that got slashed and the other 50% will get just burn and I disappear for an entire supply so what happens to the peer who committed a safety fall so that peer were actually lose in the worst case there's some like particularities in there but in the worst case of reverse the security deposits of the entire cycle cycles like well the way they measure time integers and that is purport the amount its security parses is proportional to how large you are how lies your stake imbalance is so the larger you are let's say then the more you get / basically and the live nerves is actually not penalized directly it's simply if you like your your time to produce a blog or signup for our blog comes and you're offline for whatever reasons then you were simply not win the rewards but that were not influenced like you feature likelihood of being elected again or as such in cosmos BPOs pond the perverse take in order to become a validating node and it's you rank top 100 by economic stake so and its have enough of cell phone plus delegation to rank on top 100 the way your sign slots is very so much attention in practice which is just simply the higher monitors technical proportion of a network the high likelihood are you being the one who we can elected to propose next block where is interesting in tournament or well sorry in cosmos is that they are introducing this they're experimenting with the idea of adaptive inflation and inflation in interest is like a fixed suffice of about 550 one person and only fixed by the way they introduced the rewards and buy blocks and such but in course was the looking into let's make it adaptive depending on how much is globally at stake so let's say a hundred percent of the network is of the total supply of atoms are at stake then devalue the information with tens were seven percent but if the global amount of stake is towards tends to zero percent then we do have the the fresher will be capped twenty percent but we tend towards that value so the higher value is it something interesting that teachers did not consider so in there's also similar system for like you can denounce other people but in this case in what words in the case that tells you the one who denounced God 50% of the rewards as a reward here although you did not say fifty percent will be burned but the other fifty percent would just be scattered across then try validator set equally so for slashing this is very interesting so in turn I mean the prioritized safety a lot so one safety fall would get the entire five percent of the entire stake imbalance slashed five percent is both includes delegation so I like I my own I might become valid & Kosmos and have my own atoms at stake but I might receive the delegations as well but in the case of commit have safe default the entire thing will be indistinctly slashed and there's another thing which is you get executed so you cannot become ever so your keys that one is a used for this validator you cannot use them ever again they kind of pop up ever again is to the validating set so you drop out of consensus forever you just have to generate a new set of kicks that's it and that the delegate errs will have to figure out basically who else delegate or maybe never ever delegate I don't know so there's another thing which is different from tasers is that there's like slightly harsher punishment for life nerves so let's say I haven't produced a signature as a valid run cosmos for the last 24 hours in that case I will get jail el means that you drop out of consensus but you can come back again but you will have to drop out of con you drop our conscience for seven days which means that you're missing out on the rewards you will receive otherwise so polka dot is in NPS is still under construction this is just some some features that I thought were interesting to share that were different on the other networks but again there's a lot of moving pieces in the M POS so like better ask the researchers and see what's going on so I think that I thought was really interesting in M POS is the fact that in in cosmos and integers it is the task of the delegator through the research and I choose which validators they should like delegates you right and then it's like their responsibilities to do so Pokhara dance it comes and says I think what we're gonna do is we're gonna build an algorithm that that's this automatically to maximize for the global amount of tokens at stake so when you u.s. delegate what you do is you actually submit preferences so you say let's say I prefer this for validators and then also this is like my risk scoring versus like reward scoring and then then algorithm will figure out and the purpose is to actually have a uniform uniformly weighted set of validators which in the case of you know bps and LPS everything's fantastic so like people who are more stay have more running power another thing they're saying is is that not only balladeers can submit the evidences if you are not a validator but you are monitoring the network and you realize hey this - this no it doesn't seem to be like it seems like the cheated you can also submit a transaction and you can claim her words in return just like it may be an improvement and then another thing about it is that up until then they were normally in amperes they're thinking of the idea of getting the validators token / first before getting the ones in the delegator / which in the other networks is like it's indistinct intelligence knowledge delegations are not take but in cosmos validators and alligators is equally equally punished they also in looking into differentiating between what is like where correlated falls and what are uncorrelated phones so the thing is if you watch a ders talk yesterday there's actually a lot of fun is actually quite easy at the moment so if you don't have the not to like the right tours so come in safety force accidentally because like something around managing notes and then you had a like automatic failover system something you limp redundant systems and all sudden accidentally your sponsor to see notes that were valid in the same keys and again double sign right so pokhara is coming and saying if it's a it's like evidence that there's uncorrelated so this might happen you accidentally do you but there's no thing that it was like according to that attack then we just get slashed one one one percent which is meaningful you're not so people for people to question you're like your security right or your let's say you manage your skills on on the validation service but not that large as or we're in cosmos with the entire five percent gets I lost but in the case that there is evidence there's a third of the Valladares set is coordinating safety fault then the entirety of the entire state of your state will be slashed for the entire set of like third people correlating Lee Community Safety Falls interesting to you like looking sue another thing I realized is when I was making especially this table is so tears was actually build as they designed the system around 2014-15 so they actually did not include that all of 80 years they came out active cosmos hopefully will go live next week but if you look and see like you progress a polka dot is not ready yet right if you progress a long time you will see that the system's growing complexity right they're trying to look into different things that they will consider before and they try to say okay this is where you're trying to do you but maybe we should do it this way because otherwise they another side effect just uh just a flexion ahead so what is better I don't know and I'm not I can't tell like white is better I think what matters is that which is something that I think some communities are kind of a little bit oblivious about it is that there's a lot of platforms out there and everyone's gonna tell you you know we're gonna comparing the other ones and ours is better because of that and this but I think if you're if your application that even if your application developer you have your own part you're responsible for your project you're like you will have to use this and I think it's worth actually try to take a little bit the object if you and do a little bit of research and actually seeing which platforms bit more on like the core value of my audio application so let's say security is highly important so maybe it is you're willing to sacrifice a little bit liveness for safety and then choose maybe then in that case congress of colors is better or like maybe that's not the case maybe your parrot I just maybe you're okay with not having final finalization of blocks and then just assume that risk but they still have always ongoing in transactions right so I can't say but they are here are some platforms they can look in soon here are some platforms that come with different trade-offs and I think it's worth looking into it so some challenges for prefers take researchers time for saying is that so I think that perverse take is the again like not not like in perform work perverse take is our area of research where you can actually add a lot of complexity to systems and I think what happens a lot is that there's a big difference between what in theory works and what happens in practice and that discrepancy can come from things like you did not consider all the edge cases right or you did not people often forget that feeling Brockton's is harder than writing papers but that's actually true Berkshires actually build a boxer from scratch it's actually very complex tasks and you have to think about a lot things especially in prover state which is like new area of research right but I think what's most risky to do as a protocol designer or researcher would be to simply be oblivious to the sentence of validators so simply not considering them and say you know they will figure out like whatever because in purpose thick validators are in charge of the security of network it's like if I want to compromise the network I want to start I'm going to start attacking the validators because they are in charge of security network right so you need to understand delay is the ticket to can only be like who are they why I did doing this this is purely for profit do they care about safety at all and like how are they funded these kind of things they may seem trivial but they're not because you have to think about this long-term survival of your network so just a little bit for if I thought and another thing is they want a reminder of you is the goal of perverse fake is to become an actual substitution of tufa work and you want to do it in a way where you actually so the goal is to not trade off some externalities like environmental impact or like unnecessary use of scarce resource electricity for security but it's still a lot of work to be done in both research and implementation and I would not I generally would say that a lot of these networks that were launched they were actually not be perfect in the very beginning then there's still a lot of room for researchers and people to work on this together so I think I don't have time for questions they have to okay I don't have time for issues but I'll be out here if you have any and I just catch me after thank you [Applause]