The Design and Evolution of io_uring by Jens Axboe @tigerbeetledb
The Design and Evolution of io_uring by Jens Axboe  @tigerbeetledb
Uploaded August 2026 | Updated November 2026, 14 hours ago
For years, asynchronous I/O on Linux was a promise the kernel couldn't keep - the native AIO interface only worked under narrow conditions and blocked everywhere else. io_uring is the story of fixing that: a shared ring-buffer model that lets applications and the kernel cooperate instead of interrupting each other. It's also the great unifier, collapsing fifty years of regular files and sockets doing I/O two different ways into one. The application submits an IO operation, and gets informed upon its completion. That's it. And this is the origin story of io_uring - how it came to be, and what it's able to do today.

00:36 Introduction
02:28 The world we came from
03:43 About me
04:16 Traxs - io_uring goes racing
05:00 The storage-only status quo
05:37 Threadlets and fibrils
08:04 The Linus opening
09:25 The design
19:50 io_uring and task_work
33:30 The security story
39:16 What's next

Talk from Systems Distributed '26: systemsdistributed.com
https://x.com/axboe

Slides: tigerbeetle.github.io/slides/sd/2026/Jens%20Axboe/talk.html#1
The Design and Evolution of io_uring by Jens Axboe1000x: The Power of an Interface for Performance by Joran Dirk GreefEpisode 096: Losing To Tree Of Losers Again!Episode 101: Is The Primary Dead?Episode 043: The End of the Compaction (Finally!)Ep. 17: Wiring Together the Rust and Zig Build SystemsIronBeetleState of the (Tagged) Union Address by Andrew Kelley
TigerBeetle |

The Design and Evolution of io_uring by Jens Axboe

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER