Fluence Labs Wasm in Web3, 14 Oct 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Syrus Akbary – Wasmer for web3 apps

Description

Speaker: Syrus Akbary, founder & CEO at Wasmer.io
Topic: Wasmer for web3 apps

Wasmer.io https://wasmer.io/
Wasm in Web3 workshop https://avive.github.io/wasm_on_the_blockchain_2021/#/

A

Hello: everyone I'm I'm uh sirus and today we are chatting about webassembly in web3. I'm basically like it's going to be a storyteller on how wasmer can help your uh your ecosystem, especially if you are on the centralized platform web3 world, so presentation about myself, I'm sirus, I'm the founder and ceo of westmore.

A

If you don't know what wasmery is, uh we are a server-side websome runtime that basically allows anyone from companies to enterprises to blockchain applications, basically to anyone uh leveraging on webassembly to run any kind of software anywhere in the in the cloud in the edge blockchains, and you can find reference to uh to myself and and to the company like in almost all the slides. So my email is serious that was meredith or twitter is was mariah and now our github is also was mariah.

A

So that's a bit about myself, I'm from spain. So that's why I have an accent but uh but yeah I was gonna say that feel free to ask questions. But I think this is only one way, so I'm not sure if I'm gonna be able to see your questions. So, let's start the agenda for today today we are going to see why webassembly is important, especially for uh for web3. What are the challenges that, like you, usually face, especially if you're trying to execute well assembly on on blockchains?

A

uh How was more help can help on those, and one of the things where I wanted to to put more emphasis today is what is the state of the art right now regarding the ecosystem and what the future entails? What are the next things that are coming in the in develop three well assembly world?

A

So, as pro probably like, most of you know, webassam is emerging as the universal engine for universal, but for smart contracts.

A

There are right now, like multiple engines, especially on web assembly, that make that possible, and I'm gonna review like what are why, first, why was zombies key for the web, 3 for decentralized applications, blockchains and so on? First, is because it's a universal format that can run code universally.

A

That means running code in any chipset in arc64 or arm in intel chipsets in risk five. When the future comes or any operating system in linux windows, ios android, it doesn't matter also one of the main advantages of webassembly is it can be fully sunboxed and I'm saying it can be because it depends on you, but by default is secure by default.

A

So unless we create a system, calls that are insecure web assembly and the works on the runtime will be fully secure, and one of the main advantages that we have on webassembly is because it is a very light abstraction on top of chipset instructions.

A

It's really really fast around, so we get almost uh to native speeds, usually like with a slow down of like five to ten percent and sometimes even with with the speed up. So what are the challenges when running? Well, something watery?

A

These are things that probably you already know if you are using or executing bob assembly on your on your blockchain, which is you can have a git bombs and what is a jeep on a jitbomb is a program that takes longer to compile than it takes to execute and basically like with this, uh you can have a malicious actor that tries to get your blockchain down by sending you like code.

A

That is gonna have a very big complexity when compiling so, basically, it might have like all in a square or n-cube timings when, when compiling the module- and we don't want that- also one of the challenges that we have when trying to run webassembly across that region's infrastructure is that we want deterministic execution.

A

So that means, if I'm trying to execute a module, that exit that those operations on dance and nones are not a number, so I kind of like infinity or so on. So when we are doing operations over these, these numbers, or or yeah, basically like global assembly, doesn't assure that they're mistaken execution across chipsets, and usually what we want is that the deterministic execution.

A

So that's one of the biggest challenges and that's why, like in some runtimes, usually they disable floats, which are like what usually introduce non-deterministic behavior.

A

Another challenge that you might have when running robots on this server side is that you want to scale as much as possible your web assembly modules, so you can run as many as smart contracts as possible within your chain. So these are the main challenges when using global assembly in web 3..

A

So now I'm going to be presenting wasmer here is the the marketing or my bias, my biased approach or recommendation, which is, was marisa, really an ideal runtime to execute webassembly on on decentralized platforms.

A

Why is it good for for your ecosystem? First, one is because we have a compiler that allows you to compile um webassembly code in a way that is completely linear.

A

So that means that, like no matter, if your module like weights, one megabyte or 10, it's always gonna, compile at the same speed, which is usually like five megabits per second. Obviously, like depends on on your uh on your mileage and and how powerful your chip is, but, like the compilation is fully linear. That means that, like you, are not going to see o and square complexity when compiling, which usually means that you're going to get into compiling 13 kind of programs much faster than other optimizing.

A

Compilers such as cranley or lvm on one side, I have to say green leaf is not really like an optimizing compiler, or at least it's not at the level of lodm, but uh we have seen some cases where crane lead, for example, takes one minute to compile and single pass takes just a few seconds. So so in general, like usually you, you want uh a compiler and a runtime that is not prone to jit bombs and for that westmere is really good.

A

The other thing that we have is that their mistake execution across chipsets and platforms. So that means that when you run a web assembly module in windows, it's going to run exactly the same way. It's going to give you exactly the same input same output for the same input. So this is quite important and we have like a lot of tests assuring that and one of the things that we have been like very focused on since the beginning is to have the platform to be fully highly scalable.

A

So, on that front uh yeah we have optimized a lot. The startup time is just a few microseconds and um we offer the developer and the company or the blockchain like different alternatives to to play with to ensure that you get the maximum performance.

A

So, let's continue with who is using wasner. So we got near protocol cosmosome and space mesh. This one are, I want to kind of like highlight those, especially because they are sponsors of the open source project. They are collaborating with um a sponsorship to make sure that, like the wasmer evolve at a healthy place and also to make sure that we have features that they may need, but apart from that from them, there are tons of other blockchains that are very using us to put some examples here.

A

We have uh fluent slabs holochain uh chainsafe around rit, which are just like trial and and the bad protocol. So so these are some of the blockchain ecosystems that are using us, but not all actually like. I discover each like a lot of new ones each each week and uh and modules like this one are the most most well-known ones.

A

So what we did with the sponsorships, uh I'm I'm here trying to attract you to to collaborate with the ecosystem. If you believe that, like uh web assembly is a key part of your infrastructure on the centralized world, so we optimize for startup speed, we lower it uh from a few uh milliseconds to just some microseconds, so that was uh that was a high improvement.

A

uh We optimized for really fast completion times. So in the case of single pass, we did like a 4x improvement on the speed uh when compiling, of course, if you are use parallel processes um or threads, um and one of the things that we did lately is making wasmer fully usable in the browser.

A

So what this means, if you have a ras program that is using osmorando, you could before you were only able to use your ras program in native acceptance. Basically, like you compile your program natively, you try to execute it and then like.

A

Basically, it wasn't a runs under good one of the nicest things that we did in washington, 2.1, which is not yet released, but it will very soon is that we allow your ras forum to compile dual assembly through wasn't binding without any syntax change, so you can with when using wasman in your ras program, you can target the web or you can target the server.

A

It really doesn't matter- and this is quite big, especially for what we believe is gonna make users use, uh was marine a lot of like new places, so what is coming next, regarding was america. Runtime first is thanks to the sponsorship of custom. Wasn't we have been working on full arm support? We are still on the way on developing this one, but uh yeah we are. We are getting like very interesting results right now and we'll say like we are like 30 40 percent on there.

A

Other thing that I'm super happy to to announce is that, as of today, we just merge supports for single passing windows, I'm not sure if aviv from a station s knows no resistor yet or not, but basically like. We added like full full supports for single pass, and that means that when using single passing windows, uh you can yeah. Basically you can execute whatever single pass, which is awesome. Other thing that, like uh actually near protocol has been working with uh is on adding like signal less support.

A

This support right now is quite experimental and we have like not merged yet, but it's on the way of uh basically making it work and trying to get the findings also, we want to have very easy usable, unpluggable, webassembly modules from different languages, which means we want to add support for interface types, and we are actively working on this.

A

Also. We are working on new cross chipset architecture, which is basically like refactoring single pass to make it much more scalable across different architectures.

A

So, let's look into the cool things: what is the state of the art? Where are we at right now, and this is not regarding westmere, but regarding web assembly. What are the different cool things that are happening on the happening? I mean on the ecosystem.

A

First, one. Let's talk about the sdks, I've been watching uh what fluency is with um with their framework, which I forgot their name, um and it's not very uh different from like what, depending, for example, is doing not have it. um So, basically, what I'm gonna say is probably that applies the same for for the framework that the fluence labs has been working on so with uh definitive candidate, which I think is a very original approach. It allows you to create universal sdks and that can run on the unity internet computer.

A

This is basically like more or less a precursor to interface types which allows you to create universal apis that can run in any language. I just remember what was the name of the of the framework from fluence, which was marine, so dfinity candidate, I think, is very uh similar to to marine. But there is one thing that I think is very interesting about candid, which is is formula formally verified, which I think is quite important, especially for for uh blockchain and blockchain or decentralized applications and also is unique about hashable.

A

So I would recommend you to take a look on on the project. They have support for multiple bindings and I think it's just the beginning. I probably like this is gonna like feedback into what interface types is going to become, but can be easier, is a pretty pretty good approach, I'm probably like uh um the one from fluence as well. Although I didn't, I didn't have as much time to to review it yep.

A

I think I deleted the wrong the running slides in here. I was supposed to talk about about the compiler, and I remember so. I'm just just bear with me about what I'm gonna say. So we have discovered a very novel approach for compiling webassembly code back to machine code in a way that is super super fast.

A

So right now like with single pass, we are generating the code at runtime and we are detecting like what are all the all the different web assembly operations and we are generating the machine code related to this work assembly operations at runtime.

A

This is great, but we can get much better when generating the code at runtime and as well to generate code that is more or less optimized. So, by using a novel approach that relies on the lvm compiler, we can get lvm to do that ahead of time for us for each of the box and the operations. So then the compiler only needs to patch the location of the of the addresses, and that's it I'm sorry about the slides. I will upload them later and it will have this slide fixed.

A

So one of the things that I remain super excited is what's gonna happen in the future, not just like. Where are we at right now? What are? But what are the interesting paths that we will see in the future? So here are some of the some of the ones that I remain most excited about.

A

First, one is universal: smart constructs. I think that, thanks to interface types, we can allow developers to create smart contracts in any language, so basically like marine with fluence dfinity, with candid and uh and interface types in general are going to be really key to create universal sdks that allows us to run smart contracts cross train chain.

A

This, of course, will depend on on the blockchains to decide on common sdks that can be usable, but I think there is a very good and potential opportunity for almost all the chains to to agree on a api that is going to be universal or also smart contracts. I think this might take some time, but I remain very excited about this this one, especially because when you create a smart contract, you might be able to execute in one in one chain or the other.

A

Other other feature that I remain very excited about is pause and resume. So what this means is being able to pause.

A

Our virtual machine that system will assembly module and restore it any time after so why this might become important as a smart contract get more powerful and we can do more things with them.

A

We might need to execute the smart contracts that are becoming more and more complex, so, for example, in the case of the infinity computer computer, they are starting to allow, I believe, like some passive modules move us, but basically like you are gonna get more contracts that are more complex and complex over time, and it might become very interesting to be able to pause.

A

The smart contract perhaps like when, when a note becomes like very overbooked and being able to translate or to move that execution towards other node or perhaps just kind of like being able to recover state from a partially executed contract, because something happened on the node. So I think this feature is super super exciting. We are looking forward to supporting our bosnia and I think it's it's. It's gonna help a lot on the blockchain ecosystem as well or what three ecosystems- and here is one one of the ones one of the future.

A

Things that I think are gonna happen that I really remain super excited about which is ipfs, so I'm gonna do a demo, not a demo that I have created, but I'm gonna do a demo in a few minutes on like what I'm.

A

What this this thing commenting, but basically like the the main um thing that this uh that this slide is trying to entail is that we can use ipfs as the distributed file system for both distributing the contracts or the applications that we want to execute and also to distribute the storage that this application might use.

A

So sometimes, applications like such as rust applications they might not require any like file system, but if we try to run like more complex examples like let's say, python, python relies in a lot of like files or system files that are actually required to be able to run around them successfully and leveraging my ipfs to be able to run this. This module across basically like anywhere, I think, is super super interesting.

A

So in a few minutes we are going to see an example of how, for example, run a decentralized wikipedia search, and this is going to be appeared in the browser. But it's going to showcase like what are the different opportunities on space so bear with me. I'm gonna try to share my my whole screen in a second and let's see if I, if I get it right.

A

A

So I'm gonna share my stream again, but I'm gonna share other screen. So here we are here. We are probably you're seeing you're seeing my my browser, you will like what I'm gonna demonstrate is a full thirst text search on wikipedia and that means kind of like a very large database like gigabytes of database, and here we are gonna, try to execute a search on this data set, which is super super large.

A

So let's try to search for westmere, I'm just curious. How like what's gonna happen, cool, so ah it's it's funny, uh not sure what are all these different, these different articles, but we can see that, like wasmer got searched in a lot of yeah got appeared in a lot of like different different articles, let's actually like search for web assembly here.

A

So we can see that, like we.

A

We're able to execute the logic in my computer, but the storage is somewhere else. So, let's inspect our like what is what is happening here underneath it gonna search again, so you can see the requests, let's, let's reload.

A

Yeah so basically like what is happening here is luckily, the module is requesting kind of, like the very big that data set and usually like what happens when you search on databases, is you don't get the whole database in memory? Usually, like you search at the beginning, they indices the index to see like where certain words might appear and then based on different indexes. You do like some.

A

You stick that file to get into that specific content, so you say there might be an index saying like. Oh. This word appears in this in this offset in this other offset, and this other offset and then basically, like you search on all of them, you get more indexes in this indices and then, like you, kind of like the engine, merge them together.

A

So in this case there has been like 28 requests that like come when we try to search for horace slackhorn, and this is basically like what he's doing is retrieving the storage from somewhere else, but it's executing locally.

A

So this is one of the examples that I think is going to become very key, for example, to have like search on ipfs the wikipedia, but not just that, but also to be able to run like, for example, python, but where the file system is completely distributed.

A

So that's one of the examples that I'm most excited about, and this is so far kind of like uh as far as as we can get on the presentation, I'm gonna jump into the last slide. I'm not sure if it's worth for me to share or not, but I'm gonna, I'm gonna, do it um so yeah. Just basically like that was the presentation it's a little bit shorter than the rest.

A

I wanted to keep it lean short and and basically focus more on what are the exciting things that I think are going to happen in the future. So, having said that, does anyone have any questions.