LLRT — Low Latency JavaScript Runtime for Serverless

Introduction

LLRT is an experimental JavaScript runtime built by AWS Labs specifically for serverless workloads. By using QuickJS instead of V8 as its engine, LLRT achieves significantly faster cold start times and lower memory consumption than standard Node.js Lambda runtimes, making it ideal for latency-sensitive functions.

What LLRT Does

• Provides a JavaScript runtime optimized for AWS Lambda with cold starts up to 10x faster than Node.js
• Uses QuickJS engine compiled to native code for minimal startup overhead
• Supports a practical subset of Node.js APIs including fetch, crypto, path, and streams
• Ships as a single static binary under 3 MB, keeping Lambda deployment packages small
• Includes built-in AWS SDK v3 client support for common services like S3, DynamoDB, and SQS

Architecture Overview

LLRT is written in Rust and embeds a customized QuickJS engine. Unlike V8-based runtimes that rely on JIT compilation, QuickJS interprets bytecode directly, trading peak throughput for dramatically lower startup latency. LLRT pre-bundles common AWS SDK clients so they load from native code rather than parsing JavaScript modules at init time. The runtime implements a subset of the Web Platform APIs and Node.js built-ins needed for typical serverless functions.

Self-Hosting & Configuration

• Download pre-built binaries from GitHub Releases for Linux x86_64 or arm64
• Deploy as an AWS Lambda custom runtime layer or include the binary directly in your function package
• Bundle your handler code with esbuild targeting es2020 and external AWS SDK imports
• Set the Lambda handler to your function entry point; LLRT picks it up automatically
• Use environment variables like LLRT_LOG to control runtime logging verbosity

Key Features

• Cold start times as low as single-digit milliseconds for simple Lambda functions
• Memory footprint under 10 MB for typical workloads compared to 50+ MB for Node.js
• Built-in AWS SDK v3 clients for DynamoDB, S3, SQS, SNS, Secrets Manager, and more
• Support for standard Web APIs including fetch, URL, TextEncoder, and crypto.subtle
• Single static binary with no external dependencies

Comparison with Similar Tools

• Node.js Lambda — full V8 runtime with JIT; LLRT trades peak throughput for faster cold starts
• Bun — general-purpose fast runtime; LLRT is purpose-built for serverless with smaller binary size
• Deno — secure runtime with TypeScript support; LLRT focuses on minimal Lambda startup time
• AWS Lambda SnapStart — JVM snapshot approach; LLRT achieves similar goals for JavaScript natively
• Firecracker — microVM technology; LLRT optimizes the runtime layer above the VM

FAQ

Q: Can LLRT run any Node.js application? A: No. LLRT supports a practical subset of Node.js APIs. Complex applications using native addons or uncommon built-in modules may need modifications.

Q: Does LLRT support TypeScript? A: Not directly. Bundle your TypeScript with esbuild or similar tools before deploying to LLRT.

Q: Is LLRT production-ready? A: LLRT is marked as experimental by AWS. It is suitable for latency-sensitive functions but should be evaluated against your specific workload requirements.

Q: How much faster is LLRT compared to Node.js on Lambda? A: AWS benchmarks show cold starts up to 10x faster and memory usage 2-3x lower, though results vary by workload.

Sources

• https://github.com/awslabs/llrt
• https://aws.amazon.com/blogs/opensource/llrt-a-low-latency-runtime-for-javascript/