Whiteboard Series with NEAR | Ep: 16 Qi Zhou from Quarkchain |

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:03

   hi everybody this is Ilya and Alex from near protocol and we're here with Chi from quark chain to talk about a little bit more sharding protocols right please to introduce yourself and then we can dive in deeper yeah hi hi so first of all it's my great pleasure to be here to introduce our kwok-che and also many thanks for Nearpod co-hosting this event and so coaching is next generation block sharing Network aims to offer better

2. 0:36

   trade off between scalability decentralization and also security so we already are launched our test nuts I mean public text now four months ago and our maintenance is planned to launch at the end of this month awesome okay so before I introduce our maybe detail CUF sensors probably I would like to release it a little bit how we will the province of block chains and probably we've I'm not sure new wheel point or maybe some existing but make sure that we are

3. 1:14

   solving the correct harm before we dive into the solutions okay so I think first report everybody here knows we have actually ma which is security decentralization and scalability scalar scalability and in a tree trimmer says that is very hard to achieve these three properties for at the same time but our observation of facts is that actually achieving one of them is also difficult lets us start from security first so for

4. 1:57

   example let's consider most I mean mature our concerns was proof of work which has been running for more than a decade since the deeper the beauty of Bitcoin and it hasn't it has been running for 10 years without being a little bit attack it has a well-defined security model which says this is called 51% attack which says an attacker as long as it has more than 51% of the Hodge power is able to revert any block in the blockchain and so it can double span the transaction with this

5. 2:35

   reverted blocks and so this but one seems to be a large number if the network is power is decentralized enough but the problem actually we saw is that these days for example last year become girl was being attacked with a beam depth of as many attack and also if they're classic that was also being attack in January my and that the reason is that I think there's a huge misunderstanding of this 51% of a tag especially is says 15 percent of what

6. 3:18

   people say all 50 percent of the network yes if there's a single network there's 51 % but with we say of the miners that runs the thing else it runs the same hash our zone runs the same hash hey LG and this number actually in reality is far lower so for example Eastern classic theory are using the a little outdated data where is on in January I think 2016 Easter we're steering has 176 Terra hash per second and it's 10 o'clock they have

7. 4:15

   9 terahertz per second if we just consider is if they're in class itself than 51% I'm requires five point four point five are malicious attackers but if we consider the whole miners that can also move their hash power almost instantly from other Network like you see there we only need nine points six hash power which is roughly 5.2 percent to provide a possibly attack well technically we need five point yeah it's actually four point six right out of 176

8. 4:53

   which is even less okay you can just like literally print different blocks oh no you don't need meant because this line is almost yeah we have nine all these are honest I just moved here from here right so I just moved my from here to here and proponent of spin attack and this actually what is always happened for and this can be even that the miners that can effectively contribute the hash power to network that means all the GPUs

9. 5:31

   that mines simultaneously my kryptonite echo hash right what's to my and see all the algorithm they are ready all this dimension of the hash power is can be potentially used to attack yeah I mean my cryptid 51 dota girl has a pretty good stats on how much it costs to attack any network yes and that's another topic I can say but to conclude this I will say this 51% attack is very misunderstanding because in reality even for his theorem consider all this he can be far lower than this number and

10. 6:10

    the situation become even worse with the marketplace of hash power that is fixed on a nice hash and so I can rent those hash power from nice hash and then perform an attack and the renting price according to the website you just just described to perform one our attack on instant classic Just Cause 4000 yes it's dropped to cure and let's consider 30 per hour confirmation we've averaged 15 seconds block in will for exchange to confirm a deposit that will take only

11. 6:51

    450 seconds right that is less than 8 minutes 10 minutes and which means that I can just using maybe less than 1,000 bucks on those dollars and it provides this kind of attack and which happen exactly what happened and we so we see that actually it's theorem classic and if the beacon go they are pretty large critic currency in the world like roughly top service in the world and

12. 7:22

    this implies for all the rest of POWs networks it can be easily attack and Bitcoin user is there a fee is still secure but still the rest of them are in danger and and it's not a number of attack but rather the cost of performance heart attack more related to absolute value of Hajj power this is one key observation me we see ya that we agree on there's even bigger problems if you w

13. 8:04

    but yeah we can keep going okay now the centralization I mean an interesting thing is that actually theorem after the that the hash power reduced to 100 minus 36 so this am i guessing we're out of this forties has power terror has where they move in centralization so a lot people say oh one key property of blockchain is decentralization it's very wonderful but what is the exact definition of

14. 8:50

    decentralization maybe I can give some rough definition first say oh there's no central entity to control the network no central entity oh maybe everybody everybody could join the network oh joy or contribute the network I would say is easily how to define this so anybody have question of my rough definition of decentralization or maybe better let's break them and from my viewpoint maybe quantities this

15. 9:39

    means I think that - I mean numbers I think we describe how decentralized network is I think one is the cost of running or not a maintained lady's ledger and second is the cost of producing a block so first make sure that I have the latest ledger so I know suppose a senator - and I can know how many block confirmation so that I know whether is to finalize or not and and then it can also will index server to do some specific things with might pick

16. 10:23

    some of the apps or something else and the second is that make sure that I'm able to effectively contribute to the network but for example contribute my hash power to the or maybe stake to the network and so let's just discuss this tool so the cost of Runyan nodes actually right now both for Bitcoin exterior they just using simple commodity PC fix I'm a single core dual-core CPU may be

17. 10:56

    probably one gigabyte memory weaves and I know some people doing some optimization how to use one key demand buys memory to run Bitcoin and probably the major cause is the the change size which is theorem I think the latest data is roughly 2.2 terabytes and Bitcoin I think is roughly I forgot you're engaged yeah yeah 200 300 300 yeah this number so say well say 100 gigabytes and that is still affordable

18. 11:35

    for most high paying users so that's great mmm I mean depends critical users right but you wanna run it on the phone it's easy to store two point two terabytes it's very hard to them load two possible terabytes I mean if the you know over the net also my laptop does I have two terabytes of space but but they do still affordable to say a lot of flight users maybe for example we can't easily run

19. 12:07

    thousands of nodes like for example is there and have 8,000 nodes and Bitcoin has roughly close to 10,000 nodes right that's the Sun something but the cost of producing a blog actually I should add some condition we Dowe consider powa case because you have most data on thee without joining a mining pool right if joining my name who then it doesn't help because it's centralized so in what situation and what's the cost can have a

20. 12:48

    minor that just directly contribute to the network without joining a meaningful and here we need some empirical data which miss away some of the miners to see in what situation you will contribute to the network directory or my wisdom money in pool at the price of the fee and they say if the hash power is greater than 0.1 percent which is one over one thousand of the hash power of

21. 13:19

    the network then they were likely just directly contributed because country in the mining pool basics the amortize the reward over time but suppose it had one half one 1000 which means that for Bitcoin like every two weeks it had produced two thousand sixty blocks way which means that every week it can produce one block and he and the miner is happy if that right okay so let me just take this number and see what's the cost if as a

22. 13:49

    miner own this hatch power so back to if their numbers 160 Terrace per second divided by 1000 then we have had then this mine on moves you have this one gig hash per second and now we just consider a graphics car 1080 which roughly generates 30 mega hash per second suing in like a thousand and so and suppose this one the cheapest price I can find is roughly maybe $300 and this number can easily exceed 1 million yes

23. 14:29

    not happening and then this is an investment and then you get like two ether every oh that was Bitcoin blocks obviously yet not not with GPU you're not gonna mind what will GPU will get more blocks yeah the GPU maybe relax but but still like see it's supposed to use this number then we see this is still considerable but also we know right that remaining pools if they collude yeah there's 93 mining pools and etherium that control more than 60% of all the

24. 15:03

    powers I can actually cut out anybody like given if you have point one percent they can without they lose less than I what is it like 100s of like less than cool tenth of a percent is they just cut out you know all your blocks like if you have 0.1% the mining pool can just blacklist your address and never built on top of your blocks and so your blocks will be permanently it's currently like blacklist attack right yeah all this blacklist

25. 15:35

    yep like this money like the improve of work right now it's pretty much proof of Authority there's three authorities in both aetherium and in Bitcoin that control the network arguably proof of stake is also 432 do argue but but it has more more 30 yeah so I think here I think the cost which with the thing is here this is the cause of this producer block I mean the interesting question is like when you say the centralization

26. 16:08

    right like the value of distant relation is permissionless nests right and and this like yes this is not yeah this is very permission so yeah it's permission to those who can buy 5,000 GPUs cool so scalability it's calibrating no yeah I want scalability we have no empirical data right oh yeah we do what we do nice let's see actually a lot of data is flow as the three paper okay

27. 16:46

    scalability so why blockchain is so slow because compared to centralized entity there are a lot of things are completely different so first blockchain in general have found sort of notes like this theorem has a thousand and Bitcoin has roughly 10,000 and so blockchain I was a public blockchain pu b and this will be centralized and so first and then because I mean we know need to have maintained a full lecture that means it

28. 17:34

    has very high replication which is ten thousand or a thousand Network in centralized were a hurry depends on how many like data centers some to it replication but I would say roughly ten that's pretty high number and the network probably blocking using Internet which is slow which is asynchronous I can send any message and I won't expect a return and it has low super and latency and in the centralized wall

29. 18:12

    Internet I have the dedicated data centers which have data taken dedicate wire to connect these data centers sometimes like we can use QoS guarantee this this traffic and deliver in some millisecond and bandwidth can be extremely high can modern tank gigabytes or even more in some data centers where Internet people just have around 10 megabytes or 100 megabyte megabits per second

30. 18:43

    and so you see any other numbers probably this and so if we for example in a chain increase for example the block size might reduce block interval then we will propagate these these blocks within this slow Network of all these nodes will take significant amount of time for example there's paper published in 2015 I think the name is called on decentralization of bitcoins and sometimes like that it's summarized that for Bitcoin with 1 megabytes it

31. 19:25

    will take two point four minutes to propagate a note to to propagate the blog to 90 notes with one megabyte how big is the block today let's do one megabyte I thought it was 12 seconds no no no that core in 10 minutes no no it was 12 seconds to propagate the block that's very here you I thought the theorem blog post was referring to Bitcoin yeah if it has much much smaller block size smaller blocks and they have uncles

32. 19:55

    which kind of solve some of those problems it you can call ghost algorithm but partially implemented but my understanding was that the original ghost paper ghost blog post was saying that Kahn and Bitcoin it takes 12 seconds to get to 90% and that's why you need 10 minutes to wait for the blocks and and then they introduced uncles - oh yeah this is just the data from the paper maybe a little bit faster than this because internet improve out of three four five years and 50 I've called

33. 20:34

    it this number but let's just use it is number and since this report actually is this kind of constants so which implies they're supposed to have a beacon cache which using the maximum make it besides which is 32 megabytes and suppose their notes sighs it's the thing that the Bitcoin because right now it's much smaller and this means that it will take 32 times to 0.4 which is roughly 1 hour

34. 21:02

    however I think this 2.4 minutes it comes from the fact that in Bitcoin when you receive a block you validated before you there's number of cops not the size as much well but also in practice you can actually as you receive the block you can propagate it before you're validated I'm not sure the paper at this optimization or not this kind of optimization that makes the broadcast as

35. 21:27

    long as a bypass the header check and then you will pour across immediately because as long as the typically richest and is high probability that the block is valid right but in general I mean it's about hops more than about the size at this point yeah it's hops but but at the same time when I suppose like some mysterious generally connected to 2 to 20 nodes or sorry notes for each node and I need to

36. 21:51

    broadcast block to all these things I still need me by my local panic right so need to broadcast them and in there were multiple hops and in broadcast broadcast and so and this is the empirical data to to say why why we cannot like clapping it's very hard to optimize this number definitely for have more nose than the number can be even worse ok so we see that ok why it's low because we have this network constrain especially and his

37. 22:25

    latency contains sensitive dice from one node to block us to arbitrary probably note it is it takes time and this time can create the political fog or still block and which means that suppose we have 10 percent of blocks still that means only ninety percent of the hash power is contributed to the selected chain which means that the attacker can just using 45 percent of hash power to attack network and this makes further

38. 22:58

    security issue okay so this is basic to reveal of what we saw this three problem but also I mean you're using size as kind of may weight of the block which I think in smart contract platforms kind of moving towards more how much computation actually it takes in like Bend which is not as I mean it's important part but not everything but that's correct actually when we discuss our approach to Bitcoin it's a bit man there's a

39. 23:32

    actually from like beacon network the the main cost is still network and it's not the CPU because you can process it turns out each block very fast but in theory network processing smart counter transactions takes extra time and that's why if you're a top very small abroad in who are talking this ghost algorithm and speed at the progress so that the CPU can be also better utilize otherwise suppose I send one megabytes live smart

40. 24:02

    colonist to be would take what maybe several seconds to process this smart contract yeah nice cool so let's let's see how quark chain so solves all of that yeah let's go to our solution or maybe I'll have trade off so we see this like there's a security problem decentralization program and scalability problem and and see there are actually two ways to adjust this to a problem one way is to add a blur large block size straightforward like Oh shorten the block interval and I'd like

41. 24:49

    to pick some litecoin like decreased block time from 10 minutes to 5 minutes cleaning 4 times bitcoin cash improve this number to 32 megabytes and beacon cache Satoshi vision spider increase this number to if my memory is correct one key 2801 gig what's yeah something big something even bigger I think it tried to Ames to improved why do you guys probably right now I still won today and is Suren adopted skull ghost

42. 25:26

    algorithm basically try to catch by include those ankles or still blocks into the rewards but it actually partially implement because the difficulty of this uncles are not counted to burst relative companies which means this do you have hash power loss problem no else I mean what complex is doing you can also just include those blocks in your full range rate and then speed up anymore like if you already have those uncle blocks in your chain

43. 25:59

    you actually can use them to include the transactions in those blocks that were not included in main chain as well but yes I think but still I think people have a calculation like a little off topic like hard cost I'm not sure so so basically calculation saying hey days a basically block block rate by using cost algorithm but there's also called transaction rate because there's overlap of these transactions in these blocks so

44. 26:37

    the block rate can be higher but the transaction rates still have different slope so I think that some people have to some study on that I'm not sure these two related I mean yeah it's related like you can also adopt some technique to sample transactions better too if you know that all this blocks will be included anyway that that is of topic is another solution which I think we call maybe kind of naive version of Shari is

45. 27:11

    using multiple block chains right and so we have supposed to be had 10,000 of 100 block chains it's your damn running maybe it's like a theorem like it's this EITC live aid industry but then definitely we quickly enter hatch power dilution problem which we can find chain with only 1 or 100 power ok is the target I'm so say hey it's possible that we are able to combine this these two and then come up with a solution that maybe better adjust apartment and this

46. 27:44

    is how the upper window suppose is we have a two layer blockchain the first layer this can't still can't recall shadi blockchain is that a list of the block chains that process none OLAP transactions or me now OLAP means there's a other space there's partition these transactions or senders or recipients to different chains make sure I like ghost like these have all actions actions and then to address this court

47. 28:22

    called a losing issue there's a called root chain and should the root chain itself does not contain any transactions but it's basically it has hash pointer to prove blocks as normal blockchain its body to it's at least half hash pointers to the to the shot block chains and as long as this body consists of valid sequence obsolete each a then this route block is valid and so on a short chain the fork

48. 29:12

    choice rule will always favor the latest block which appears on the route chain right that's cool let's say someone created a longer chain right like from here they created a longer chain yes but for as long as this block is here and this one is not this one will be preferred but we always prefer and actually in this case even for this case because for this one is corresponding ledger the full adder is not including

49. 29:42

    this one because this is corresponding to this one the cost of a full ledger is this one and Plus this one so it's corresponding route changes to this one so even for this one's do this one so no matter how long this is that the attack will always fail we call routine first for choice rule so the good thing of this is that for any block that is confirmed by the root block it would be automatically protected by the rule block and suppose the rubra has

50. 30:13

    significant amount of hash power like bitcoin is theorem the mission is that all the shadi blockchain they can be protected by the better rule block and this is basic key idea but but here's the the question right so let's say so let's say we have like you know five people or like five five gigahertz are mining this chain right now let's say like five gigahertz mining this here

51. 30:43

    chain and let's say we have 170 terror terror hashes mining this chain so this chain is significantly more secure but but the particular validator here right so let's say these a particular machine that is mining he write that his machine cannot validate all the all the blocks right so effectively when when someone minds a block over here and they and they send it to the root chain the root chain all

52. 31:14

    that it knows is that five gigahertz is attesting to the validity of the block but but they don't know that the block is valid right so if I'm corrupting this five gigahertz I can produce a block which is completely invalid how to corrupt this five years i I bring ten more gigahertz of malicious so I I myself control ten gigahertz which is a very small percentage of this right yes I bring ten

53. 31:37

    gigahertz for you you can only create a new block right and I create a new block ah but this book is completely invalid so let's say it but other all the rest notes will reject this ten the invalid block but no no but these ten gears you can just produce your own chain right my my chain is longer right so how does have this root chain so let's say I created this invalid block right maybe

54. 32:02

    maybe what is invalid block another is may be attacking fog along the fog or some but much longer I can see how this protects against longer right but let's say so this is the this is the so five gigahertz I will use green for honest okay so five gigahertz is good people thank you get hurt is something I rented on nice cash that one nice cash and I created the block and in this block this block is completely invalid

55. 32:27

    I just in this book I just said the email is like victim I can mind additional coin or maybe okay and then obviously this five gigahertz eight all this block is invalid we're going to continue building this blockchain and they continue building on this blockchain but I am faster right so I as a malicious actor I'm building a longer chain and so both of us sending the blocks to the root chain right how does this particular person on the root chain know which one to snapshot into the routine block that comes to our

56. 33:02

    definition of how we can't decentralize in this room we are going to adjust so basic there are two problems one is closed off running on note and second is closed off producing block so you see that running unknown is me belief is more bottleneck than running a producing a block is the more bottleneck of runyan note so we assume the network but no actually is running the full edger kind

57. 33:29

    of attack are you saying this person they actually have a note for every red shirt yeah but the point is that we have been called a cluster design which allows and you can imagine a load or super node or a cluster that is able to distribute all this short tasks form its sub machines this is we can't call if you said if there is the world computer then we say we are with the world cloud right and because we see compared to

58. 34:00

    this the cause like contribute is directly the running an old cost is much much less and suppose we just need to spare maybe 10,000 bucks to maintain this no but a lot more people to join Network then we see we believe we are cheating some kind of motive centralization that exists and 10000 is per shirt or 10000 you thing is the price of a full cluster but is it full cluster or is it one

59. 34:28

    shard is it highly depends on the future TPS of the network because this charge can be added right if the blower is small value there were more people to run the node and probably this will become may be higher than production and we will like to see that but when you say 10000 does it refer to the cost like how many shards do you have in your mind in mind when you right now my laptop can run like easily run a shorts

60. 34:57

    yes and see if if I run we have our TAS net that is able to run 256 shorts I think that will cost how much we have 50 nodes I need to check chocolate that's but 15 notes is one cluster 8 so you have a cluster which can continue ticking costume 50 clusters each cluster runs to 129 notes with one runs for Uche and 120 a runs for two shots and a niche and each node is a separate machine effectively separate machine yes Oh separate

61. 35:45

    processes using tcp/ip eyes and so Justin for me to fully understand so one cluster is an abstract term for one or multiple entities running one process per shard right see if you have 128 charts a cluster would be 128 you said it was - it can be arbitrary combination you can be one processes run multiple shorts but at least one process one one shot and one root chain so one cluster is one

62. 36:25

    independent snapshot of a chain one independent snaps of processing entity yeah I mean his fullness full full ledger of the other chain and you would expect mostly one cluster to be control fully by single entity right you can be single okay okay over control my friends my trusty and it is so I mean that's pretty much a new pool it's but it's a trusted pool and there's no slashing its proof-of-work so and then which which

63. 36:56

    hash function or like what what's the proof of work and actually one problem if you can lock it can you see any consensus it doesn't need to be proof of work it can be displeased important work and use this hash hash value as kind of see to select proof of stake validators because it is so strong edge power so making treat treat treat on this from one validator and once that away that on one char is not probably not highly unworthy right you can also run perform

64. 37:29

    work with different hash algorithm actually it is similar to existing like a mix off that is there or this but with more ways so that everybody can join the hatch power of the root Campitelli about it we call kind of and any actually and in short is added in flight so we saw for example we can add a sharper year you see a one shot another year one shot and the char would be immediately be protected by the hash power and we call this called hash power reuse technique it's kind of kind of tell you reuse

65. 38:04

    because you stand there libraries why don't we use same some standard hash maybe pull up something to patate new chains I mean the question is pretty much yeah if I'm mining the root chain pretty much makes sense for me to run this because if this is like 170 Terra hashes I need to invest a million dollars in GPUs it may as well run this but if I'm right mining one of the shark chains right I will not

66. 38:30

    be doing this right depends for example let me back to previous equation so for this suppose I have a sharps that's just using some numbers a shorts and oh I forgot there's a how how this let's just use a specter like suppose all of them using the same hash power this probably the simplest case and so we have a color tax mechanism so that some portion because I'm a 2/3 of the block rewards - we work

67. 39:02

    with allocate - row chain and one third will allocate to issue all the shots in that situation then the the cost of joining the pool will be original number maybe 1 megabytes per million divided by 8 hey sharts and divided by 3 which i think is close to this number right so as long as the cost of Ronnie Ono is smaller than this month number then it's my question is why do you need hash

68. 39:42

    power here at all why don't just say anybody can produce blocks was like no like everybody who's checking in them in the root chain will validate them anyway yes so why not just produce blocks they check them in valid say correct block like what why do we need any hash power here whatsoever like why not just have block producers destroying the blocks being the block B so just just include

69. 40:07

    all the blocks by or even better the root chain block for the root chain runners because they all have any all the chains anyway running they may as well produce all the blocks that essentially actually so we have this code tax rate tax to adjust this hash power between them the base essentially it is this has no hash power this is no incentive and this is one write all the incentive is all

70. 40:32

    equated to here so that nobody tried my it no no but like why would I would yeah let's say I'm running this cluster and I'm I have you know million dollars of GPUs to produce blocks and on the root chain yeah why not I will just produce all the blocks all the charges but that would be effectively you've if even if it's one third I'll get all the one third as well because I produce the blocks I'll produce the blocks and

71. 40:59

    include them into into the next rule chart change but but but but they if it's one third here and two thirds here those who you need to do the work separately well I mean five okay like this is so much no no no in that case this will be divided by 2 so this will be 1 7 Tara hash / 8 / - right this is actually hash power of this otherwise it will call because this is market 66 -

72. 41:30

    Tara hash queue right 16 Tara hash well I mean my question is why do you separate them at all why not just like if you already have this class test running yes right what is the value of having like some of the people producing well I guess someone has to propose blocks yeah but but this folks can propose blocks very efficiently because they have the full state of the chain of

73. 41:53

    all the chains but how do you choose the block that is like simultaneously 1,000 locks are proposed from 1,000 clusters which one do you choose but you're the root you you proposing a root chain lock right these blocks into produce significantly faster than the root chain root chain is like once a minute this is once every 5 seconds you know yeah maybe 10 10 seconds so I mean you can still produce like if you are running this

74. 42:17

    cluster you can be producing that's where what happens you have one third of has power here just just to order them in some way yeah and then you snapshot it yourself with blocks so probably I can maybe I can explain this way suppose that we may have no incentive of producing a block here so that has no hash power so basically the rubric can't just include whatever it wants and this is

75. 42:40

    essentially the problem that the same solution has large block size right essentially the same and suppose we have no hash no hash power on the row chain and remove routine first consensus and distribute all the hash power to all the shot block because it's essentially the same as multiple project that's why we adjust this tax aims to harvest both benefits especially why we do not move

76. 43:07

    audience because when we propagate this large block as I mentioned before it is latency sensitive but if we have multiple blocks produced block simultaneously and asynchronously because because this guy is block this country is block so originally the network usage of Bitcoin you can imagine like this there's a block receive after roughly ten companies are not gonna block receive there's a limit Charlie four receiving transactions and there's

77. 43:38

    another receive right so it's just latency sensitive but a lot of bandwidth are not being used so right now we have another you can imagine each one is basically a described but with summation so that another one may be like this like this like this like this another Chaffey is like this and with summation of this we won't disparage to be like like this yep yeah that makes sense I think it

78. 44:13

    sounds similar to Bitcoin ng that's like an older paper I think from I mean yeah I think because because you desks or some inspiration from them but I think it it just have one of this and understand like this doesn't have it circumstances but now we extend this to if it's come sensors and for example I can have one of them using the depot's and so for the users that would like to have instant finality and to make and

79. 44:46

    then we can also have another one will be runs different lecture or virtual machine fixed angle maybe one window windows type of and then we can build kind of enclosure transactions to make sure I can say maybe if they're like tokens to da middle window Chang and do the privacy research and Sascha on that and move back and all these things is decentralized I I mean in my definition

80. 45:15

    okay so yeah how do you do cross your transactions in this case yeah that's good question so basically let's just consider a simple cross transaction pilot example parents transfer for smart contra we can also apply similar principles but so for balance balance transfer basically just choose up operations why is withdraw money from sauce balance and then deposit the balance to destination right and unfortunately in cross Shore transaction there's this has to be done in a

81. 45:49

    synchronous way because in in charge and Sasha I just done within a block and synchronously and so basically there are two couple two major problems so one how to guarantee it is secure because suppose I the money is or a deposit to the destination sure while the original shall be enough spent there how to make sure that the money will not be fixed on my team with extra supply and another is how to make sure this

82. 46:22

    transaction deposit will be being included what is called was the token economy all incentive to one destination shall like to include I just I don't care right so first question autumn st let's just let me see yeah probably just simplify the graph like this let's just let's just assume there's a transaction here that is going to send from from zero so one concentration to shanwa and

83. 47:04

    here in our design actually for every sharp rocks besides it has a hash pointer point the previous block it also has a posh pointer point to previous rubric a hunt point 0.02 block the sum of this can be disrobe lock and based on their best observation it does not guarantee to observe maybe the block just just in front of it it can be you can dip the block before each block but

84. 47:40

    as long as this one because so we have a lot of the same thing in our code and so for this one for so for this block suppose it will not process this kasha transaction until it observed the rule block that contains this block this bit because this basically becomes a partial order right and with routine first consensus now back to that if I want to revert this block which means I want to

85. 48:20

    also revert this block which will also invalidate this block I guess you know and this make sure we always guarantee this and our relationship if these two sub I would say sub operations okay this is first and felicity and second is regarding the incentives so for each block it has two gas limits unlike if they're if they were have a we call may be called local gas limit another coke or chorus gas limit the cost could gas limit is doing what

86. 49:01

    actually and it is it is for each block header it has a cursor which tells what is the the last cross-cultural transaction that is and the block is being executed that fixed time at the cursor because this contains a list of this there's this blocks way this contains so it contains with a cursor this says oh the last block a process disclosure transaction of corresponding this block and the transaction ID there for each of the other and and the route

87. 49:45

    chain contains all the crush our transactions no it just contains that they had the hash pointers but it will process this or the cur the cursor points to the particular block and at a particular transaction within the block at index just index index of the transaction yes wait so this chart you need to receive the full block no actually when the block is being received it will broadcast all the costs transaction to the other Shores or if in

88. 50:18

    a classroom this is classic I would say let me echo such as close a transaction design so that because if I put in a rubric and definitely ruin others so it will basically broadcast to all the time so once this one observe this block all the other corresponding cause or transaction of the block included but it is rule block is already observed by this guy store in a local minimum PO you know and

89. 50:50

    there's a very specific order in which croushore transactions are executed trade it depends on the order in the block in the in the in this in this block and then the order is included in the Rue block so route blocks actually order all of the other sharks blocks in some fool order you must be for the otherwise it will include about invalid shot blocks right because it is quite a canonical chain of all the shot chains but not just that it also across all the

90. 51:22

    canonical across all the shark chains what is the order between them as well because like how do I know should I be executing cross shard from here first uh from here right yeah so suppose there's a block here that's also included as well so this route block with first public force included yeah so it's it's a list it's not per shard it's actually a full list it's a list that the in humanity's list

91. 51:45

    yeah it sounds like shard one block 23 shark to block 27 shard one block 24 it's a return so it's a full order we also have defined this neighbor relationship just in case we have a lot of shorts maybe thousands of chars directive direct communication of this will cause this cause a transaction communication cost to be n square which is number of shorts and so if we fix this number so this will be M square and x number of the number of shorts so this

92. 52:24

    is total cost and then you'll just like route the trunk like they will be rerouted I can imagine these like t-shirts yeah like let's say these for short so this is gonna block this is shorts I should have used some other symbol and let's say they connected like this and I want to send transaction from here this is source this is destination right so I will just go I will not send it directly

93. 52:47

    I will I will write it right so effectively so this block this short sorry they will produce a blog that will include this transaction and they will do nothing with it they will just they will just include the but there's a lil bit early it cause to be in our experiment even to 10024 shorts we still pick the class that can city of process disco efficiently all these are 256 I haven't asked 110 24 but 256 which has maybe

94. 53:16

    it's 7% of course a transaction we haven't see any performance degradation so once we reach that stage there may be considered its proper but when we said oh looks like this is another only problem we want to solve at this moment another way is that we have to just call wallet even so basically this is automatic routing right so they say I tell sauce destination and your routing and we can also use the wallet to do

95. 53:41

    some manual routing by issue multiple construction sashes but that will be probably annoying to users but at least at a current moment fix motors near 12 456 and test different work percentage of course our chance - and I haven't seen any performance degradation there okay so one question here is so let's say there's a local gas and you said there's a crush on gas how does this short

96. 54:10

    actually get the crusher gas so well it process this one by one right it will consume the Kolchak so when you set when you like send a transaction here you actually deduct whatever much money plus I mean plus the fee just the remote feet pretty much yes but the fee is spent I think because you cannot you cannot get directly here right yeah but so for payments it make sense right but let's

97. 54:36

    say it's a smart contract you don't know yet how much gas will be spent on the second charge right I know because the stock has always is specified by the initial transaction I see oh you just subtract the maximum gas specify I haven't yet but like in a theorem for example I specify how much gas but I get refunded yeah so so there's the refund but we found thanks to our addresses space

98. 54:59

    every every I would say I was every 160-bit we can't public key in theory address recipie you has corresponding add has corresponding edges in all chains so in this case we won't return this remaining to back to this account because we feel this is make concessions as a more efficient it would just refer to a local account by changing I would say we are necessary to bits like like IP address to indicate which drives basic operating and sorry we return to the local and because we

99. 55:39

    see this is very small amount and we have this small wallet who able to collect all this if use of file out there too many balances different charts and they want to clap back but so if everybody has accounts like from sending money if everybody has accounts on all the shards like I can just send it on on the same chart right most of the times that's my contracts different yes so first my car wreck sure that's polluting isn't it yeah and actually in

100. 56:12

     ours design the cost like some etherion is doing a course a transaction is roughly with start with disk an initial instructions on hands discuss right so we add another 9000 transaction this is a basic kind of the cost gas cost for internal transfer right because we treat this kind of like it runs a big smart contract with internal transfer and so actually the cost between close a

101. 56:43

     transaction in a transaction I feel is it's not so significant yeah so the user took somebody to give me a jazz either one but maybe I thought well to send my local shop because I want to stay save a tiny transaction fee independent how much transaction fees are insistent on yeah that depends on gas pipes are the gas price right so this times gas price can you cover everything yeah awesome okay well thanks a lot please check out Courchesne wipe

102. 57:30

     every test net and leave comments on youtube thank you