High Performance Node.js Powered by Rust and WebAssembly

Hello, everyone. I'm Shivai Lamba. I'm currently TensorFlow.js special interest group member and also a working group member from India. And I'm also a mentor for TensorFlow, specifically for TensorFlow.js.

And today my lightning talk at Node.js Congress 2022, it's going to be on high performance Node.js powered by Rust and WebAssembly. Now first, let's briefly talk about the Moore's law. Everyone knows that the Moore's law stated that the number of transistors on a microchip doubles every two years. But if you sort of see from the previous past five years, the size limitation that has caused, has sort of limited the amount of transistors that can be set in a chip. And that has sort of resulted in a plating effect of the performance of a computer hardware. So that means is that in order to be able to overcome this limitation done by the hardware, we have to improve the software with the help of practices of improving software performance. And at the same time, these software improvement performances have to also be safe and secure at the same time.

So for that, we are going to be looking at a few tools that you can use with a lot of different server-side applications and languages such as NodeJS and Rust. Now, why to choose Rust? We have a lot of legacy programming languages like C++, but these are specifically error-prone. But Rust provides a safe, modern and is also equally, if not the same as C++ in terms of efficiency. And also, it is much more capable than NodeJS in terms of performance. You can really crank it up with a lot of highly demanding performance applications that can run very simply with the help of Rust.

And then, the question arises that how can we actually use Rust to code on the web? Well, the older way was to actually compile it to the native libraries and extensions with existing runtimes like NodeJS. But this approach proved to be actually quite unsafe and also required a high amount of maintenance. So, what's the better way? Well, you can use WebAssembly, which is essentially a way to be able to... it provides the instructions to be able to use bytecodes that help interact with the application that you're running it on, specifically the web, with the lower-level applications that you're running on the platform. And we are able to just compile the Rust code into bytecode and then manage it with the help of a virtual machine such as WebAssembly. And what are some of the benefits of WebAssembly virtual machine, which today is also known as Wasmage's runtime? Well, it provides a lot of security out of the box. It is extremely lightweight and portable. It provides extreme efficiency because it uses techniques such as ahead-of-time compilation. It also provides runtime safety as well. And it's specifically optimized for a lot of different server-side applications. And that sort of brings us to the ecosystem that we have for this talk. That we want to use JavaScript for its ease of use to be able to build applications while also have the performance of Rust and have the safety and portability of something like WebAssembly. And that sort of brings us to this architecture that we have. So generally, Node.js runtimes will be based on C++-based runtime.

But of course, we have the alternative solution to be able to essentially have a WebAssembly-based virtual machine which can run the Rust-based programs that have the highly computational tasks in them and that can help to make Node.js-based applications highly efficient. And let's see how that actually works. So what we're doing is that we have our Node.js application where all the highly intensive tasks are not written in Node.js but they are actually written in Rust. So with the help of a JavaScript function, we'll be calling the WebAssembly-based functions that help to bundle up the Rust function and compile it into a bytecode that is then served and is actually hosted on the WebAssembly virtual machine, in this case, Wasmx or the SSVM. And that helps to not only secure the application but also all these highly intensive competitions that need to take place are now being done with the help of Rust. And you just have to simply call this function with the help of your Node.js environment. And of course you can use them for a lot of different things, including AI inference, because you're able to directly actually communicate with the native hardware that is your AI hardware or you can use them for things like storage databases as well since WebAssembly is highly optimized for server-side applications.