Ethereum & WebAssembly

Transcript

it was today I'm going to talk about virtual machines woz I'm short for web assembly but I'm not going to just talk about web assembly and just give a little overview of virtual machines all together I think virtual machines are pretty cool okay so forth um like really the virtual machine is pretty core to all of it so like what do we want from a virtual machine architect right we want to we want really good performance we wanted to be close to the metal we don't want a lot of abstraction between what I virtual machine is doing what the real hardware is doing yet we also want it to be portable we wanted to run on many

devices so we still need a layer between like x86 and made of assembly and our virtual machine we also would like it to be standardized right we don't we don't need to recreate the real another really nice thing would be to have like tool chain compatibility it would be nice to use llvm to the compile stuff to run on Virtual Machine and lastly would be good to have a clean and fairly simple instruction set or is a so yeah sorry x86 so to back up a little bit a traditional or normal hardware we

have the execution environment a CPU house like an execution environment has been any parts but execution environment and then it has a core where the ISA the instructions actual opcodes get executed it and the ISA is usually pretty simple I mean it doesn't there's no instruction for example to load something from the hard drive the way that is handled is you have physical addresses and they write to memory and then your CPU just reads from memory reads in right from

memory and that's the memory sections are allocated you know a boot time for the physical devices so we sort of have an analogy to this novel arm so traditional traditional CPUs they keep it isolated the instruction set is modular it doesn't necessarily it isn't making the assumption about the environment that is operating in right it doesn't make assumption about what kind of devices you have plugged into your computer right no op codes for a maps or whatnot so this is what this is what etherium is current is a it looks

like so this would be the intrinsic set we have memory operations system operations so things that give us information about the state of the virtual machine flow control and we have plenty of integer operations and then going back to the analogy of the traditional CPU these opcodes would be like external devices so I want to draw electro three and gas call and precompiled now those are some interesting and i'll talk to

talk to him about later but so like block hash and code size and all these things these are properties of an external environment that need to go into the virtual machine but through a modular interface I'm necessarily having the opcode form so the rationale putting the call lock code in with the external external operations that might not necessarily be clear but our co-op code is not is not like syscall like if you

ever done unix programming or something you have a special just call to interact with the OS our call our call is a little bit different we can think of etherium so gap has saying what it is like the theorems like a singleton virtual machine which is really nice but the I think of it sort of like a singleton hypervisor and each single contract is its own vm or like its own vm in the sense that it has its own storage space it has its own its own isolated environment that it works in so going back calls like calls aren't

syscalls in traditional sense when you're doing operating system Paul's armel are like a vm talking to another vm um so I think it can be external okay uh gasps so gas is a property of the virtual machine and if you had pure hardware that analog to gas would just be you know clock cycles um so that would be nice oh so but really found out something really interesting actually not all CPUs have clock cycles I anyways there's actually it's not too relevant

here but there's actually some some CPUs that are totally asynchronous so you can't you can't seem that all hardware has clock cycles and then lastly a precompiled contracts they're a necessary hack that we had to do right because for performance um but if you had a very performant virtual machine and you didn't have to rewrite you no chalk 3 and AH sha-256 and ec recovered you could just compile it and run it on the Virtual Machine you wouldn't need the this pre-compiled contract so the end result here is we have a stronger vm

we can also have a more consistent interface so what would alternative you know possible virtual machine look architecture you look like so if you want to look at hardware a really nice one today is risk 5 so there's multiple risk stands for reduced struction set so um there's multiple variations of risk and risk 5s I think one of the newest ones that has a lot of momentum so yeah so is a it has open source open source it's open

source is a so the specification soca music all the people developing it also include an open source implementation plement ations in it so the hardware is the rocket Corps the implements is a so it's just like little teen court and then the whole CPU is there's one implementation low risk so there's multiple implementation stupid ones called low risk and it uses the rocket court so yeah so the good thing about this is like it's really close to the metal right is metal and it's standardized that's great how's

toolchain pit batch ability you can compile any C or C++ program and you can matter fact just run linux on it it has a very clean design and but the bad thing is i mean if you wanted to actually burn one of these things to chip it's very expensive so you probably wouldn't it published emulate it development with it is fairly tricky it wasn't necessarily designed it was designed for hardware people wasn't designed for software people if you're not burning it to chip you're emulating it you gotta have to modify it some and wasn't really designed for that in mind

environmental info information so like loading external devices like the blockchain stuff there's no there's no strict way or set path to do that it's assuming you're loading stuff into memory and then reading addresses from memory or reading information from memory and that's really unnecessary because in etherium we're not plugging in pulling out devices all the time so there's really no point to have that part picture and lastly it's not necessarily portable as once again notes yeah designed for hardware so another alternative might be web assembly the

basic idea behind web assembly is its assembly for the web basically JavaScript wasn't fast enough for certain applications so people realize that ok we need a little bit lower level low level abstract than something really high javascript so they came up with web assembly actually first the history was first they came up with a SI MJS from Mozilla and which was a subset of JavaScript that made it really easy for interpreters to see sections of code in tournament age it and then it was really performant I was kind of ugly another alternative from google was portable

bean apple which allows you to run compile wildest in e c++ code and run it in a safe virtual machine so that's the sort of the predecessor to those with sir the predecessor to web sembly and the current version of web assembly is a new is a that is designed to be pretty ultra really fairly low-level portable and fast and also safe for web browsers strong it is a result of the collaboration between Google Mozilla Apple and Microsoft which Chris you know we have doing all the major browsers and

replace it snap a line sometimes yes another interesting thing about it is even though it is fairly low-level it does have some higher-level abstracts to it so the instructions are specified specify some AST abstract tree semantics so for example functions modules and has variables instead of registers but it they're really interchangeable so it's a registry-based we say variables and it has it's really easy to represent a national expressions which is commerce so yeah okay three minutes of that so you have some AST semantics which we're

going to skip the ISA in the instruction set so structure set is pretty I mean okay so what if people didn't sign an instruction says for like a pretty long time right now and they really haven't changed for the past 50 years so no big surprise in this director said right it's a 8-bit bites two's complement in a juror little ending is comes in two variants 32 byte 64 byte bit variant so why would web assembly possibly good for using on a blockchain well it's built to be performance it's built age it on built to be jit it so at the right so like it would came from JavaScript which

is like really hard and dynamic digit is it's basically like a compliment to something that is really dynamic but really not so dynamic easy to it opcode compression is built in compressions like arm pretty pretty important on the web when you're sending rights across the wire so that's even more important in a box chain it's portable you can should be easy to run on many different architects it already has llvm back end meaning you can compile anything you want to that lvm can compile to it and target it this specification has a

pretty clear specification and they're working on a formal specification when okay home and it's built to be run in a hostile environment so what we do need to change to modify web assembly to work on a long hanging with him like environment it needs to be fully just deterministic so web similarity does a pretty good job of isolating non-deterministic behavior um so that would be pretty easy to you all you have to do is disable threads and a few operations that uses approximations that might be hardware-based you have to add access to environmental information so

like getting the block half and getting a different environmental information from block chain and you'd have to add me drink that's a pretty big one um so yeah this is the environmental information you'd have to add as i said earlier web simply has the idea of modules which makes it really nice to add all this environment information in since you could implement it as a module good implemented in a module then that would have different functions to query the box chain state and that would that model would hook into the action late of

code so there's a clean path of how to embed this instruction set into external environments so also know the big thing is you need backwards compatibility and it turns out trans piling evm code Moslems pretty straightforward i like to do in two different stages and is this not complete project but what you do is you would first translate it wazzle in you would translate it to awesome plus metering some metering code that would run on a normal awesome machine but if you modified was them so that you

had the metering built-in natively you wouldn't do the final step so that's why it's nice to have it compiled like that and here's some images of how transpiring looks but my time is up so all right but thank you