Layer 1 Blockchain Architecture (SFBW)

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

1. 0:03

   okay so we have so we're going to

2. 1:10

   program this thing over the summer Shepherds gonna be great didn't didn't work out it turns out it takes a longer than that so it's been like over two years now we still been like building and we've just gotten like a public testing that we like over the same watch-chain with the whole consensus later in the sermon turns out there's a lot to do there's not like one big thing but like once you go between that you know like there's like every part of a computer science stack almost have to touch like

3. 1:41

   build something with cryptocurrency they put a label on vertical and by the time you touch Sharla starts to realize to hear some like well what are the different words see a distributed system is you networking you have like photography you have like particularly in issues and Julie consensus algorithms which is like you know a very complicated topic in itself you have data structures you have like you know

4. 2:07

   systems level engineering cases and like just yeah so we started building gently in January meeting about two years ready and we are going to launch our main that next month's running and we did several I think it it it takes so long to beautiful day one option because there are so many moving parts you need to you need to it depends on what kind of

5. 3:15

   if you are working on proposed a proof work you may simply succeed bomb is actually not especially when you trying to improve proof of work at the same time new hash function so yeah for decentralization and we we improve the consensus to make it more natural to network condition and you have to build a new virtual machine and when you try to well sometimes you find there some existing liability news or news for

6. 4:00

   example maybe to be using oh we can just use this memory because it is not sure and there are so many problems the library in the Ross implementation so we have to give up we we basically library yeah there's something the whole truly is not mature you need to over yourself I think and you have to combine them together to make them work yeah I think that's why I guess my taking it if you don't agree with everything around Sun

7. 4:44

   but like you know you try it when you actually convince yourself they should still the later ones they're kind of crazy and you have this vision it's gonna be the best possible way to run I'm afraid it's just going to be so much better than anything else that's out there and then admission is what they think introduces delays is it on the ship like a forked aetherium weather like it's week I mean people have done that but people people tell them but it's kind of pointless so because of

8. 5:17

   that I think like we're driven to do something that is missionary and then he realized oh Arthur Sox networking sucks there's no tools everything has to be built from scratch again and it's just a lot of work almost every morning what's harder what's hard early community stuff or the engineering so I think this election you can raise a

9. 6:25

   lot of money by publishing paper which has like some time th neither let me come here insane to build it because - of issues right so that's why you see so many protocols from that they usually show so who never shipped right if you look at the word near like a thermos existed one and a half years ago both of us a practical remain and what the college is extremely complex so so it's it's finally now to speak the winter where all these like those fake company

10. 6:58

    sometime is the way you changes the complexity but they both actually if nu equal to constant giving a little bit on but marginally and it just mirror so the

11. 8:15

    empty page moment again because wasn't exists and was slow

12. 9:24

    watching some of the properties without changes attribute our home and tell us because giving an estimate census is extremely complicated

13. 10:26

    the later one's watching very few people has experienced this field and if you hire new engineer that we circle is a great excuse more we need

14. 11:47

    one people to come together and help each other all right okay cool so I think we all sort of got the impression that we're one ball chains apart they're also fun okay I guess which is like sort of execution smart contract scripting as a bunch of different approaches here like Kota has to pull everything inside a bike it's the same functions you guys are doing your best don't you nears all

15. 12:51

    into blossom there's harmony things won't fly down okay so like maybe we talked a little bit about like approaches to code execution like what the trade-offs you see are and what why you should make the choices what what you think is important is embodied in those choices about what the head of business watcher can you put on it function sure so I think let's see what I wanted to start yes so Bobby purse Dakota Kirby net is there

16. 13:24

    will be like a small set of like template and functionality like the limited set of our conferences recalls but and this is because doing things inside the star is especially a bargain monk in this perspective of like doing general comment you can you like really good like you X so probably happened at some point but like prioritizing like tokens and things like that the thing that I think is important for us here like that higher level is it should be really easy to do like the piece things you shouldn't have to like

17. 13:55

    create like a million smart contracts if you like so simple multi state and your system for like cooking thing so supporting those would really be dressed like the most important part I'm alright before we go back when we tried to make a decision the execution model what we think about first is was the feature from day one not change we think there should be global consensus everyone yeah

18. 14:37

    and such a consensus because the scope is so large everyone change and that affect a lot of design decisions like for example we chose to use a low temperature machine instead of a high level one we knew the Newbridge machine which is called CVM adopts with five instructions and if you know it is designed for it is designed for new ships there someone said that blockchain

19. 15:47

    is made hardware software it's really like hardware once you get a design and it's out there it's really hard to change the design you just cannot ask the community to heartful as you wish that's why we that's why we chose to use reservoir instructions design specification is already finalized and the span is modernized so we can choose the module 83 1 and I don't need to worry about a future extension to the spec so everything's changed and I have too hard for my option 2

20. 16:36

    so I can benefit from the salaat ecosystems in comparison if you chose a high level virtual machine webassembly even maybe it'll be a heart for example because it's high level there will be something that garbage collection the specification and do you want to when when you want to look better send me I'm not changing you will think about it right do I need a garbage collection home workshop not if not I need to

21. 17:20

    remove that part from an assembly and that's why I think I think that's the reason why there are so many web stanley variations in this watch a movie it's not like everyone when my web sent me and you are using a different version of Medicine so yeah that sometimes we have semi like we make decisions based on was the fastest implementation physically possible really comes down to how to utilize

22. 18:01

    heartburn and the only fast target we have our GPUs or you know Sydney lands so the signal instruction multiple data execution model sub our virtual machine is based on this really old crufty bytecode built from the packet filter around since the 90s it's used in the Linux kernel to take untrusted code from the user space and send it to the kernel and then execute those PDF by cut bucks

23. 18:33

    advance in the kernel and it has some really nice properties where the again JIT it you know single mouth starts 86 or other Russian languages and it doesn't have any stats so why that's interesting is because when you have Harper that's basically running single instruction multiple data those sing some of stocks is some data pointers so you know do a push and pop on the stack and then executed instruction just run

24. 19:05

    code or a bunch of data it's our goal is to take contracts compile ones out to PDF check them to spear by running on its purity and then run out of GPUs because when you think about it what is what do you think is going to run on the blockchain right it's gonna be an event where like a centralized exchange alongside a couple crisis is a comparison to the French and then a store and it's not going to be a complex program like the website in York Times comm loads a thousand different modules

25. 19:39

    runs a bunches like images and animations and usuai interactions all on chain right all that stuff relieved gnarly branchy like business logic but should be off chain when you're doing up chain is it's just basically figuring out who owns what and those kinds of contracts are very small and very small number branches and are less perfect or that kind of executions so really our goal is to give you the best performance

26. 20:12

    per dollar for the level of security of a layer longer so so I'm gonna be depending wasn't but before there wasn't nothing that carbonation so so so there are two reasons why we use want some and I guess reason number one is that one would think about volchok image of the day and that is very different from the coaching came into several years it is that they will change image is really collaborative

27. 20:47

    right so many many projects they work together on many in many aspects of research right so for example but the moment we published some research on them before space-time space matching mediately reached out to us collaboration so project do want to collaborate and so once in is one of those has so many purchased a lot greater affinity is variable on three super awesome both other pieces possible Sweeney

28. 21:11

    reduces one are capillar the projects and so because of that many aspects of using wasn't in block chains acidic we already sold right super circle one big aspect is determinism is imagery deterministic right if we want people to reach a consensus on the outcome if the M produces non-deterministic results right so one of them is already adjusted most of the existing compiler is already it is variable what integration from them to make the deterministic specifically we use Posner

29. 21:42

    and we have all the proper changes to ultimately TVs so that's that the synthesis of program was near its opposite my task of the native code to execute as it's a it's a single it's sufficiently massive is the perfect thing but this one important thing about was it is done once it has main languages comparison

30. 23:09

    okay so the reason we still use DBMS we want to be actually by being incompatible those existing maps can easily just migrate with several single staff like maybe ten minutes you can just migrate to our change and directly enjoy all the benefits and that's actually one of the main reasons we still be compatible actually I really think you may have to save a lot of news a lot more right thank you you get to a market faster so besides

31. 24:04

    competi we're also thinking about Watson in the future because it gets more language supported and about getting more people or news of launching able to ramp up or enter Lon Chaney's easily already you don't want anybody to learn solidity when they when they were not watching they want us to use their own language problems or others yeah so um because it's synchronous right what

32. 24:34

    happens if two pointers from different sharks and these assume principle so for the water and of our wholeness award I think this is a very hard problem to solve and right now there's there's no one answer solving on all these complicated the innocence you could support and a simple operations like plumbing transfer go play yes its own youth open using the shark which is busy using the same technology of normal they did public transfer

33. 25:39

    I mean a simple one or maybe she won't located one zero building where one who particles do your layer 1 particles need a layer to succeed because we are we we need to drugs to help us to we did many specific design like children we we make our virtual machine so efficient so yeah basically and because two of us

34. 27:43

    two three so they're two different so yes higher security guarantees and the generational cost will be higher for there also some diseases of the verification they

35. 28:48

    need to be sent around what a lot knots for layer two is right it's more faster than thermal security grantees much less and in the network is your friends movie as at-risk where you need to rely on the layer one to set up and so the function linear essentially as secure as the layer one but the benefits of layer two is definitely in some of the cases where you don't care about the security so much and the tradition values oh you

36. 29:20

    know yesterday's the users don't care about whether is secure or not they just need the instant kind of payment or instant treatment between people right so this case like caving in definitely need there to support awesome