Rearchitecting Linux Storage Stack for µs Latency and High Throughput

on YouTube ↗15m

" summary

blk-switch is a Linux-kernel rearchitecture of the block layer that maps block-layer queues to driver queues and adds per-class egress queues, prioritization with preemption, and dynamic steering (request- and application-level) to avoid head‑of‑line blocking. Implemented in-kernel and evaluated over a 100Gb link against Linux and SPDK accessing remote in‑memory storage, it achieves ~10 µs average and ~166 µs p99.9 latency while sustaining near‑hardware throughput, reports up to 76× average and 11× tail latency improvements over stock Linux, and requires no app, scheduler, or NIC changes.

AI-written summary · about

" chapters (13)

1. 0:00Intro
2. 0:20Widespread belief: Linux can't achieve ps-scale latency & high throughput
3. 2:05Performance of Existing Storage Stacks Applications accessing in-memory data in remote servers (single-core case)
4. 3:42blk-switch Summary
5. 4:44blk-switch Key Insight • Observation: Today's Linux storage stack is conceptually similar to network switches!
6. 6:07A deeper dive into blk-switch architecture
7. 7:22blk-switch Prioritization
8. 8:16blk-switch Request Steering for transient loads
9. 9:24blk-switch Application Steering for persistent loads
10. 11:02blk-switch Evaluation Setup • Implemented entirely in the Linux kernel with minimal changes (LOC-928)
11. 11:43High Contention Scenario (In-memory)
12. 12:59blk-switch Performance Breakdown
13. 14:13Summary . It is possible to achieve pes-scale latency and high throughput with Linux

" related talks

• 45mLinux Storage Stack ExplainedWerner Fischer · Open Source Summit Europe 2024
• 42mUnderstanding the Linux Block Layer-From Single Queue to Multi-Queue ArchitectureNitin U. Yewale · DevConf.IN 2026
• 16mRearchitecting Buffered I/O in the Era of High-Bandwidth SSDsXiangrui Yang, Hong Jiang, Qiang Cao, Jie Yao, Yekang Zhan, Haichuan Hu, Tianze Wang, Zheng Peng, Jiahao Wu · FAST '26
• 42mUserspace Block Services Based on Ublk and Vduse in SPDKChangpeng Liu, Xiaodong Liu · KubeCon + CloudNativeCon + Open Source Summit China 2023
• 32mUsing Linux Block Integrity in Building and Testing Storage SystemsMikhail Malygin · Vault '20
• 38mRethinking Writeback: Scaling Linux Filesystem and Memory Performance for the Next DecadeKundan Kumar · Open Source Summit Europe 2025
• 17mHow advancements in storage are triggering innovations in open-source and vice-versaNitesh Shetty · IndiaFOSS 2.0
• 34mublk virtual block devices in user spaceStefano Garzarella, German Maglione · DevConf.CZ 2023