Day 2: Emin Gün Sirer (Cornell University)

Transcript

[Music] please give a big round of applause to today's keynote speaker professor amin goon serie hello thank you very much for having me here and thank you all very much for showing up at 8:30 on the second day of the conference so i want to talk to you about blockchains and i want to talk to you a little bit about the hype and where we are today and the gap in between and how we fill that gap so where are we today we're at a very interesting point in time if you were to

walk around okay here we go if we if you were to walk around lower Manhattan or Zurich and you were to observe any two bankers walking on the street you would note that they are turning to each other and asking what's your strategy for blockchain not a blockchain not the blockchain but a word without an article and the only other word that I know that's in this category is the word God so it's become quite quite a thing of its own and has taken a lifetime a life of its own cryptocurrencies have emerged as a new asset class worth more than 100 billion dollars today there are hundreds of currencies are 1600 as of last count last night about two billion dollars of

traditional VC money has gone into the field about 10 billion has been raised in the form of AI cos that's an amazing amount of crowd funded resources going into this nascent field and there is an incredibly rapid pace of development the systems themselves are very robust our Bitcoin itself has had absolutely zero people running around with pagers this is very unlike Google Facebook and so forth and despite that it has had no down time in the time that it's been up and and the meetings are themselves like this one actually bring together a very interesting mix of people you have stolid bankers we have even a minister yesterday coming together to the same table with tumblr operators and

in darknet market people and one of the things i like to do is sort of look for indicators from social media to see if there is real grassroots interest in area and one of these is an is an interesting one which is the number of songs written about a particular topic there are many more songs about Bitcoin and blockchain compared to how many songs that are about Google Facebook and even the free software movement this thing is really big so so what's really going on behind the scenes so what I think is going on is we're witnessing the emergence of a new class of systems so the back end of almost every big technological company is going to change and if that were the only thing that was

happening then that would happen in the background and you guys would not mix with my kind and my kind would do that in the background and and you all would go about your business and thinking that ok the back end is now slightly more efficient we got some efficiencies perhaps but nothing new nothing big is happening there's another second thing that's that's a foot which is we have new financial instruments that are emerging and on top of that we have with smart contracts the ability to disrupt existing business models this is an extinction level event for many companies whose job is to stand in the middle and it's going to be there many of them are going to be replaced by new modes that are enabled by smart contracts and the elimination of trust

and middlemen so this is all fantastic but but what's what stands in our path well ok so I have a new ass message to give you there is a deep field here there is an amazing opportunity on the side of Business and Finance ok it's not just yes it's not just computer science but at the same time there is a giant gap between what is being sold to the public and what is achievable today the systems that we are capable of building with blockchains today are not all that decentralized they are not that fast they do not handle as high a throughput as they should they do not scale to the levels that they need to scale to if for example all of Venezuela were to switch

to Bitcoin today every adult Venezuelan would be able to go to the store once every 36 days ok so this stuff is just not ready for primetime in its current form so we need fundamental new science to reach the what dinero is actually telling us now that path is possible and I want to give you a glimpse as to what it takes to get there but before we do that let me uh report to you some of the studies that we've done on on on decentralization in in blockchains now the full details are in various papers that my group has published but I want to talk to you about about a recent study that we did where we looked at the amount of decentralization in Bitcoin and

aetherium so so of course decentralization is crucial in this domain it's not only crucial for again reaching the narrative but it's also crucial from a legal perspective the SEC in the u.s. recently announced that the legal rules that apply are different depending on how decentralized you are so it's it's essential that we measure the level of decentralization and it's essential that we understand what's happening underneath this these systems so so of course the centralization then is this unicorn in the name of which many system design decisions are made so we did this big study to examine just how decentralized the Bitcoin and aetherium networks are some multi-year

study about two and a half years of data about five gigabytes of data collected and so on and so forth I want to give you just one simple graph maybe two simple graphs out of this whole thing we looked at where the the mining power is in Bitcoin and aetherium and these graphs over here summarize the various the various different mining concerns and their hash power over time so every collection of bars corresponds to a miner and how their hash power changed over the course of that over of our study so you can see that somebody you know sometimes lags behind then adds more a hash power goes up and so on and

so forth you will immediately notice that the sum total number of bars I'm showing you is less than 20 okay so these systems when it comes to constructing the most important data structure that ledger that that's so crucial it comes down to 20 coordinated or coordinating units there producing these the actual blocks so this graph shows the CDF of the cumulative distribution function for the same data and you can see from this one a couple of interesting interesting takeaways the most important one of course is this 50% line right there and to summarize it it's simply the case that during our study period bitcoins

blockchain was manufactured by sum total of about 19 concerns four of which could could together determine the entire contents of the the blockchain so four of them would be able to exceed the 51% boundary the numbers for aetherium are slightly worse they come down to about 11 miners and about three of them dominate and have about more than 51% of the hash power so what does this tell us well it tells us that if we could somehow put together a system that that creates a Byzantine quorum of about twenty nodes or so we would immediately be as decentralized or perhaps more decentralized than these systems up

there and in fact somebody did just that recently about a week ago heõs went online and it created a lot of fanfare raised about four billion dollars by itself from the public and it comes down to twenty one block producers in AOS okay so these are not very large numbers these are the kinds of numbers that the master students working now over two weeks could put together with the help of some very well-established algorithms and then deploy on AWS and suddenly you have something that is similar in in the number of nodes that are powering how many that are powering the the blockchain production so clearly we need we there is a big gap between the decentralization as people imagine

it and the actual decentralization on the ground and I would like to talk to you a little bit about new protocols that actually allow us to scale way beyond these kinds of puny numbers that you see here twenty one is too too too small and we need to go up to numbers that are in the thousands range and the base problem here is the protocols that we had access to just can't scale to that level and I want to give you a glimpse of a new protocol that can so I mentioned there are many other protocols of iota for example is a centralized protocol this has a centralized coordinator at the moment there are many many many copycat currencies litecoin dogecoin etc etc and all of these come down to bitcoin reaper

ammeter I see they take the Bitcoin source code they change a couple of things here and there too in the name of quote decentralization typically and then they end up having a new artifact that's essentially a replica of the old one has all of the same faults as the old one achieving decentralization and achieving low latency and high throughput at the same time has been elusive it's very difficult now my group has been pushing the envelope here and and when we actually try to change these protocols to achieve these these higher levels of throughput are the lower latencies that we need to be able to check somebody out at the at the cash register we find our two big problems the big one of the big problems that

I'll talk about next is that the network layer the network layer has often been ignored and therein lie a lot of problems on information dissemination we have a lot of financial information in the hands of people who produce them we have a lot of people who demand them and connecting the two has been often overlooked a lot of effort has gone into the crypto in cryptocurrencies and not into the network that powers them we do not currently have an Akamai for cryptocurrencies so so what can we do well okay so what we would love to do is scale the network so Bitcoin currently achieves about somewhere between three

to four transactions per second a theorem achieves somewhere around fifteen transactions per second research yves on average de about 2000 and has the ability to to repeat up to 50000 transactions per second so we need to find three orders of magnitude somewhere we need to bring that into the picture where is that going to come from well one option is to build systems that are our perhaps centralized but untrusted this is a portion of the design space that until now has been overlooked that as people have been trying to build these systems where everybody is an individual actor there's absolutely no coordination between any parties and there's

absolutely no trust between any parties what has been overlooked but I would like to explore our systems where there is a centralized party that operates a large number of nodes but is not trusted whose operation can be verified okay so this is a different kind of system design business people typically I'll provide a function and and then say trust me I will give you this so this is not that kind of design this is I'm going to give you something that you yourself can verify independent of me it'll do what it says and if it doesn't you'll have proof that it didn't so to this end we've been building this infrastructure called blocks route whose purpose is to facilitate the sharing of financial information so as I alluded to

before there typically are things like miners who produce blocks and Merchants who need these blocks to see who paid whom and vice-versa the miners themselves need the new transactions that need to go into new blocks and that connection has not been done well to date so the traditional approach in these systems is as follows when you have a source of data and appear that wants to consume that data and a bunch of nodes in the middle and let's suppose that time is going down the source ends up sending the data is let's say a block to a relay in the middle the relay receives it then checks it it takes some time to check that the data is received as good and then forwards it on so that

process of course takes some time it takes some time beta for transmitting the block and some time alpha for checking that the block is good so if you sum that up over n relays or so you're going to see that we have a latency of of about alpha well it's actually this if you have n of these relays of n times alpha plus beta that's n hops and so for every hop you're paying a hefty cost of receiving so that is storing first checking and then forwarding the day so what can we what can we do well one of the things that we learned from computer networking in the 80s was that store and forward networks inherently are slow and there's a very simple

technique that we can bring to bear through this problem so once again if we have the same setup and we have a relay in the middle that relay need not actually store and then check and then forward the block what it can do is it can start sending it as soon as it receives it so that time period from the beginning of its actually receiving the data to the time when it starts sending it can be made tiny the time it takes to make sure that this in this case the block has a properly formatted header so if you do that in at scale to over an hops you will find that and the network is a very

different characteristic instead of taking a rather hefty linear delay that is proportional to the amount of time it takes to forward a block which is also a very very large number we can actually bring it down and D couple it from how long it takes to send the block so I have some data I start streaming it and the first person to receive it starts streaming it as soon as he receives it and so on and so forth and as soon as I stop he has it and very shortly thereafter the next person has it and so forth so this is the basis of a new network a publish/subscribe network for financial information called block strand and the elimination of that linear factor gives us an amazing advantage in how fast blocks are

propagated so this allows us to to decouple the amount of time it takes to disseminate blocks from the size of the blocks themselves so that allows us to have much much larger blocks and therefore much higher height through much higher throughput systems so we've been deploying this on various different knows we have a giant Network Operations Center and we've been distributing data from people who have it to people who demand it in a in a manner that is auditable and checkable even though I'm not going to talk about those aspects of the system I want to talk about where we get our performance advantages so this is part of the work that my group did looking at

the often ignored Network layer but there is another layer up top which is the consensus protocols and the consensus protocols to date are there have been many I'm sure you've heard of all sorts of things that people have tried to sort of repackage and sell to you they all come down to fundamentally two different approaches one of them is what we call classical family is this family that has been studied for approximately 35 years or so and this was pioneered by Leslie Lamport and Barbara Liskov both of whom have Turing awards these are the classes of consensus protocols that powered permissioned permission blockchain so if

you're doing an enterprise deployment this is typically the kind of system that you are actually dealing with now these systems all work by doing essentially this suppose we want to achieve some kind of consensus in the room a closed room of this kind while these require that we know exactly who's in the room and once we know who's in the room then we can determine exactly how many people must have heard of a message before it can be committed I need to make sure that enough of you have heard of something so that that other person over there will definitely hear of that same thing when when he asks and therefore we I can commit it safely knowing that he or two will have to necessarily commit it safely so all

of these correctness proofs come down to intersection and being able to argue that even in the presence of up to fpf Byzantine participants Byzantine meaning participants who can do whatever they like the participants who lie who say one thing to him and another thing to me even in the presence of up to f participants I know that another honest person will will most definitely with absolute guarantees are going to hear about a particular transaction that I think has been finalised so you can see from my description of how these systems work that this requires a very rather fragile set up I need to know who's here if

I disagree about who exactly is in the room than those overlaps and any safety criteria based on those overlaps will not work so that's one of the reasons why when you guys do your enterprise deployments you are unable to deploy them to groups larger than you know to groups whose membership is unknown to groups whose sizes are rather large now the second problem of course is typically in these protocols what needs to happen is I need to talk to everybody and so does he and so do you and so does everybody else and that requires n-squared communication so if I were to try to use this with thousands of nodes then the system would grind to a halt because there'd be millions of messages just to

be able to commit somebody buying something at the cash register and so that is of course not an acceptable solution strategy so and that's exactly why when AOS for example yose uses a protocol in this family and of course they are limited to about 21 nodes because if they were to try to go to larger numbers then that square factor would get them now the thing that gave rise to this big revolution that all brought us all together in this room is another family of protocols found by the pseudonymous Satoshi Nakamoto person so the Nakamoto consensus family works entirely differently it says look it's

too hard to try to keep track of who exactly as part of the system it's too hard to decide who's got what it's too hard to it's it's too fragile to build systems of this kind instead what we're going to do is we're going to create this blockchain using this technique called mining where essentially we do a lottery okay so you all try to solve a crypto puzzle and somebody says bingo I've got a good solution and that person then determines the next block okay so that's the core of the Nakamoto family it's incredibly robust so another person could come into the room we wouldn't even have to care about their identity we wouldn't even know and if they come up with a block then they scream out bingo and we add a

door block to till our blocks their blocks our blockchain and everything is great so this is fantastic the problem with the Nakamoto family of protocols of course is that the latencies are very high so bitcoins latency is 10 minutes to add the next block the throughputs are very low it's one block per block interval so in the case of Bitcoin 1 megabytes per 10 minutes comes down to 3 to 4 transactions per second and so that does not scale that gives us those scalability limitations that we see and worst of all these systems are necessarily based on wasting a lot of energy all these miners are constantly trying to solve these crypto puzzles to earn their rewards and that

of course creates a problem we're consuming I think yesterday's talk alluded to this about 3 to 4 gigawatts and if you want to visualize it that's about 3 to 4 nuclear power plants or if you want to visualize it another way that's about 2 denmark's worth of electricity consumption is going into bitcoins bitcoins blockchain production that's a lot of energy and sure maybe some of it is coming from renewables and so forth but know that energy would not have gone to waste otherwise we could have diverted it into functions that are more positive for society so it would be fantastic to try to direct that energy elsewhere now if you don't buy the sustainability and green this argument

let me also mention the money argument the greed argument we have to pay these miners immense amounts of money to do that task and they're wasting all that energy keeping each other at bay so the work that they do to serialize blocks is actually quite minuscule compared to all the work that they're doing to come up with hashes that that to compete with with the other miners themselves so what happened recently well something really interesting happened about a month ago that you all should be aware of a new paper was dropped on IP FS and it outlines a completely different approach to the Consensus problem and it combines the best of these two families in a very

unique very interesting manner now what is that well this is a system that's called Avalanche and it achieves the quick finality of of Lamport discover of your classical consensus protocols it achieves the high throughput in fact higher throughput then and than any of those approximately somewhere between a thousand transactions per second to ten thousand transactions per second in the geo replicated setting it scales to tens of thousands maybe millions of nodes its robust it doesn't require precise membership like Nakamoto and it's quiet when there is no decision to be made it consumes no energy it's green it's sustainable it gets rid of mining it

gets rid of the minor ecosystem it gets rid of divisions of labor between the user community there is no difference between me and you versus a miner okay so the miners are gone the core of this is for those of you who have a distributed systems background is is an algorithm that is designed to be metastable at an algorithm that's designed not to remain in balance when it comes to consensus protocols balance is absolutely your biggest enemy you don't want the protocol to sort of be flipping and flopping and be wishing and washing between two different choices it should quickly topple over to one decision or the other so for example in a choice between two colors red versus

blue the protocol should just go to either red or blue in fact we don't care which one it fix will we very much care that it quickly picks one of them so okay so let me just try to give you a sort of a hint of how these systems work so many of you might have seen very long-winded descriptions of of classical consensus protocols you might have seen these like 60 plus page documents on proof of stake protocols in the past that are inscrutable in fact many cryptographers at the last financial cryptography conference we're sitting around the table you know scratching our heads and saying it's been 2 years of these proof of stake protocols nobody trusts any anybody

else's proof of stake protocol and somehow somehow you know any the authors of any one paper are highly critical of the authors of every other paper as have no visibility into what's going on here what's happening can't we have a simpler protocol well this is a much simpler protocol and it's something that I want to sort of you know outline for you in the space of the next five minutes or so I think all of you should be able to leave here knowing exactly how Avalanche works and what the intuition is behind it so we're going to be building a new platform based on avalanche we're going to be building a new currency called alpha coin it's going to be a payments rail a transaction rail and a computation rail

is going to support smart contracts as well as payments and it's going to support them at a scale that's unheard of with the level of decentralization that's previously unseen in cryptocurrencies and at the core of it is this new protocol so here's how it works it's very very straightforward we start in a very large crowded room or maybe let's say actually make it bigger imagine you're in a very very large Stadium two fifty to a hundred thousand people you don't necessarily know everybody in the stadium of course and in fact you and I might differ on who exactly is in the stadium and that's completely okay the way this protocol works is everybody when it's their turn they just pick up okay it's my time now oh it just looks around takes five

people at random maybe five to ten so this giant room you don't know who's there you just pick like five to ten people and you say hey between the choice between red and blue which color do you prefer okay so imagine that everybody has a color on their forehead and so you ask these people which color do you like and so they come back to you and they say okay well I like red red blue red blue and at that point you say okay it looks like there is a slight dominance of red in the network at least as far as I've sampled it and I'm going to flip my color to red myself okay so that guy changes color and it's then you know at the same time actually it's a result at the same time somebody else is doing the same thing somewhere else in

the network this other node here is sampling and he happens to also sample red in this case and you can see that after a round of this okay the following will happen the worst case scenario is we started out with a Stadium that was 50/50 evenly divided and the worst outcome would be that if the stadium would remain 50/50 divided forever but the chances of that happening are tiny because what will happen after one round is there will be a slight preponderance of either red or blue based on random choices we will end up over sampling either the red or the blue we will go from 50/50 to 5149 most

likely and after that at the second round we will likely go from 51 49 to 53 47 and you can quickly see that the network will develop more and more and more of that same color that initially ended up dominating and and then at some point in time what we call the face shift point the network will begin to move faster and faster and faster towards the solution and after a certain point in time what we call the point of no return the network is safe to commit that there might be some stragglers so I slowed but they too will most definitely end up sampling the exact same color they too will not be able to go back to

the initial bivalent state so you can see that the shape of this decision curve looks exactly like this and the chances of remaining at that apex are tiny after a modest number of rounds it turns out the math shows that after about 11 to 17 rounds in a stadium as big as 50,000 you will achieve you will look around the stadium and everybody will have independently perform this really dumb protocol really simple protocol and you look around this ginormous stadium and everybody is holding a red card on their foreheads that's an amazing result it's very very robust and it doesn't require any kind of of membership any kind of precise knowledge about who's in the system

it's a gala terian it's green and so what does that mean there is no miners there's no mining and everybody is at the same level okay you participate in the system by buying some steak you get some other coins to say okay here I am putting these coins up here to show that I am an independent in the that I am NOT a million copies of the same person so that's a useful useful feature to keep out an actor who might otherwise pretend to be a million other people and crowd out the the reasonable voices in the room so with that that stake that's set out to show that you have some skin in the game you get to have a voice in the system and with that voice you get to determine the decisions

made by the system the system overall I don't have time to go into the technical details the system overall are constructs not a blockchain but a directed acyclic graph so at the same time you're building a giant graph of transactions a paid be paid CPD that kind of a graph and that allows us to have maximal concurrency so multiple people working at the same time can be extending this giant ledger that is in the shape of a graph as opposed to a single linear structure that single linear structure has an inherent or creates an inherent bottleneck so I will be showing you some numbers that are very very fast and one of the reasons why they're fast is this optimization based on the graph but the core idea

here that made it possible was this metastability now in Silicon Valley it's sort of you know you have to seem wise and one of the ways in which people seem wise is by saying well you can't have a free lunch there always has to be a catch so for one that's false okay so that's just something that people say to to appear like they know what they're talking about it's not actually true now you can easily have protocols that dominate other protocols in every way possible yeah we've seen this many times in distributed systems now it turns out in this case there is a bit of a catch if you want to view it as a catch but it actually isn't a catch it turns out that there is one thing that this protocol cannot do which is if there is a sufficiently large attacker in the

system then the attacker is capable of issuing two transactions where Alice pays Bob and Alice pays Charlie at the same time and then the attacker is able to keep the system from deciding between those two that is every time the system is veering from that Valis point if the attacker is big enough perhaps the attacker can pull the system back this turns out not to be a problem because the only person affected in with this limitation is the attacker himself okay so it's an interesting way to sort of try to ball up all of this badness that has to go somewhere in the system and to put it on the shoulders of the misbehaving party here so there's this the the lack of liveness in the presence of an attack is

something that only affects the attacker and not honest transactions so okay so what are we doing with this what we're looking - looking forward to building a new coin this new coin is going to be called avec Oien it's going to support multiple virtual machines so there's going to be other than native token which supports very simple operations and on top of this anybody can create their own coins of any kind that they like in particular though we would like them to be able to create Alva BTC for Bitcoin or Alva etherium or Ava's e cash or alpha manera and so forth we would like to be able to support the Bitcoin script or the etherion virtual machine you know Z caches ek snarks one arrows

ring signatures and so forth there is no real reason why not as long as you have isolation set properly in your implementation there's absolutely no downside to supporting multiple virtual machines and it's very very advantageous to do so I alluded to how stake can push back on the power of an attacker - to try to mess with the system in reality though we expect that once you have a good enough reason to push back on the attacker the attacks don't even arise in the first place okay so we have seen in public deployments the attacks you don't see them at all until they appear you know towards the they appear only when when guaranteed to succeed let's see the one of the most

interesting parts of this protocol is the governance scheme so the way this protocol works which is subsampled voting subsampled questioning of the crowd to try to achieve consensus is quite amenable to polling the crowd to find out what it is that they are willing to go with and that is a the basis by which we can use the protocol to determine meta properties properties of the proto call itself so bye-bye that I'm referring to of course the one of the most important metal properties the determination of one of the most important properties and that is the interest rate so in Alva you can have transactions that say something like hey everybody I think that we are minting

too many coins we would like to pull back on the rate at which we mint or vice versa we are not minting enough coins we would like to we should we should provide more incentives to people to join the system to put up a stake and to participate so we should crank up the interest rate in contrast Satoshi had to come up with his emissions curve with his Milton curve way ahead of time in 2009 he had to fix that curve that issues a sum total of 21 million coins and he got it right sometimes and he got it very wrong some other times sometimes demand out faces the curve the price goes through the roof and you can decide whether that's good or not typically most people consider that to be bad for a stable currency and other times much worse is

demand lags and the price tanks and it would be much better to be able to not have to bake that in and it's difficult to do so so but this this approach to to protocol design allows us to do exactly that so when people talk about consensus protocols they typically have this notion in their heads that that they could that these processes protocols find and the emergent consensus that's not what happens in Bitcoin you pick a miner and then he tells you what he likes there is no notion of an emergent consensus in Bitcoin but in our there is so you can do exactly that with artha okay so this governing is going to happen through repeated sampled voting there's of course going to be some

reasonable ranges so you can't move the system to a an undesirable point of operation let's see so we would like to be able to to enable people to create their own tokens I would very much like other tokens to be able to switch to Alva so this new consensus mechanism it's green is better than in every respect then then mining based coins so if you if you are you know if for example if you are Z cash you would benefit immediately from switching to avalanche right Z cash currently is emitting way too many coins and should not be doing so and it also needs a differentiator that sets it aside from Bitcoin and other copycats let's see so this gives

us the ability to achieve unprecedented decentralization so Bitcoin and the theorem as I showed before comes down to about 2020 concerns that operate together well here now we can have thousands upon thousands of people participating in the system so there are always going to be some people who like proof of work so some people really fell in love with the idea you know I did too initially when I read the Satoshi paper and those people really liked the security that comes from pal well those people could still benefit from avalanche by layering it on top avalanche could come up with the decisions about the mempool when there's a double spend in the mempool avalanche within the course of a few seconds can make a decision as to which of the the

transactions should go into the next block and then the miners with a soft fork can enact that so if you really like your pal well okay you could still take advantage of this Nuala mechanism so let's see oh yeah I have some time so let me give you my quick rant on enterprise block chains many of you are working on proofs of concept on the dealing with these super fragile classical consensus protocols and many of you are doing it wrong let me also mention that how many of you are deploying a single monoculture of code across a modest number of nodes across a dozen nodes so that alone so the rubs me the wrong way as a distributive systems

person because these systems are supposed to fail independently and if you're running the same code on all of them I'm sorry but you have a giant single point of failure across all of them you're doing it completely wrong so but let's leave that aside maybe five to ten years from now you will have multiple implementations of the same code base or the same interface and maybe you'll get that right but I assure you that your business models are actually not very well aligned with the kinds of concerns that classical systems require if me and you are about to go into business we typically look at each other in the eye we create a bilateral relationship but with our enterprise blockchains we have a very different model you and I

are supposed to look at each other in the eye and then we go okay let's have our records be kept by Deloitte latouche and this and that PwC etc so we somehow have to then determine a Greek chorus that works on our behalf to to keep our records and and every time you and I transit we're supposed to turn through this Greek chorus and say I paid you five bucks do you guys acknowledge and they're supposed to say yes we acknowledge this is a crazy way of setting things up and no wonder very few PLC's have actually moved into production because no sane company would like to deploy in this manner if this Greek chorus today isn't charging your money well tomorrow they will because they do put them in a monopolistic situation you have to pay a rather large number of knows more than

one of you know five to ten or more for keeping your records so whatever you used to pay yourself now you have to pay fivefold to tenfold to somebody else and so that's setup I find is going to be a very problematic one and I suspect that many enterprise block chains will stumble at the gate after very successful pocs PLC's will succeed and production will not move and that's exactly what we've seen for the last few years Alvah is the right platform for these things because you can actually move the the data to such a decentralized fashion that you are not beholden to a monopoly that you and I don't have to create that fragile consortium so we also are of course going after smart contracts

that's an immense new area for us and if we get this right we have the ability to to scale them up by about a thousandfold and that of course allows us to have crypto kitties simultaneously with a very popular ICO something that has eluded the existing smart contact platforms okay let me give you just a couple of brief glimpses into some very conservative numbers so you can see here that the system is capable of achieving about 1600 transactions per second in a deployment of geographically dispersed deployment of about 2,000 nodes here is latency the very first one is Bitcoin

that's 3600 seconds okay so that's one hour if you want a really secure confirmation of the equivalent for aetherium is about thirty-seven confirmations comes out about five hundred and fifty seconds or so okay ten minutes and there is a very interesting system by a highly respected colleague of mine Sylvia McCauley at MIT and his system Algar and achieves and after using very fancy crypto in the classical context where you have to know all the participants achieves about latencies of about 50 seconds and so that's still in my opinion unusable I would not like to wait behind somebody buying coffee at 50

seconds avalanche is capable of doing this somewhere between I think this is a very conservative number of four seconds other experiments that we did range from 1.2 to 1.7 again in the gia replicated setting okay so where are we today well we stand poised to transform how business is conducted through tokenization if we entirely fail if the entire blockchain area goes away and all we have done is come up with new technologies for keeping track of of custodial financial assets then it will still have been a ginormous improvement for mankind but I think we're going to succeed you've even beyond that we're

going to succeed far beyond what anybody else has has foreseen the big stumbling blocks to this success are twofold their performance and scale and to date the focus on decentralization has kept us from achieving the performance and scale that we need to really reach this dream the narrative that has been sold is really compelling but the technology on the ground in the trenches is unable to get there so with these kinds of new technological improvements I suspect that we're beginning to finally crack this puzzle that we know what to do at the network level to get the information from the people who have it to the people who demand it very fast and to achieve consensus in a manner

that circulat Aryan green and super low latency and high throughput so with that if you have any questions I'll be in the in the area and the conference later on I'm very happy to chat with anybody who's interested in avalanche thank you very much [Applause]