LLRT: Optimizing for Speed and Cost Efficiency in Serverless Computing

9min Snip

00:00

Play full episode

Summary

Transcript

Episode notes

Exploring the development of a bespoke lightweight runtime, LLRT (Low Latency Runtime), to enhance speed and cost efficiency in serverless computing, particularly for IO-intensive tasks like data transformation. The chapter delves into the technical aspects behind the faster startup times of LLRT, incorporating Rust and QuickJS to tailor features for serverless environments.

Scott and Wes chat with Richard Davison from AWS about LLRT, a new runtime tailored specifically for Lambda. They dive into the benefits of using LLRT, challenges with JavaScript in serverless, and why Rust was chosen for its development.

Show Notes

00:00 Welcome to Syntax!
01:07 Who is Richard Davison?
05:11 What is LLRT and what’s the motivation for building it?
08:25 AWS Lambda example.
11:20 What makes LLRT specifically tailored to Lambda?
14:55 Brought to you by Sentry.io.
15:22 Node.js in Lambda.
16:00 What are some challenges that people have with JavaScript in serverless?
17:20 Lambda memory configuration.
19:23 Managing cost of compute.
21:29 Simpler and faster than Node, Bun, Dino, but not a replacement.
22:31 The benchmarks.
27:00 Quick.js, the main reason for the performance gains.
- Fabrice Bellard QuickJS.
28:03 The Quick.js engine.
30:35 What was the reason behind creating Quick.js?
33:46 What made you pick Rust for LLRT?
36:34 Abstractions and the value of speed.
39:08 The JIT Compiler.
42:38 Compile cache.
43:27 De-optimizations.
44:59 Node.js Compat, what to use and avoid with LLRT.
- GitHub AWS Labs Compatibility Chart.
47:52 Will you target with WinterCG spec?
50:22 Streams API.
52:06 What about WebSockets?
53:10 Is this going to be promoted from a labs project?
54:49 Sick Picks + Shameless Plugs.