Programming Emerging Storage Interfaces

on YouTube ↗35m

" summary

xNVMe provides a single user-space API that maps NVMe commands to libaio/io_uring or SPDK, handles device enumeration, DMA-capable buffer allocation, request pools and callbacks, and aims for high performance and high productivity. Benchmarks show io_uring hitting ~1–2M IOPS with polling, SPDK reaching higher raw IOPS, and xNVMe adds only ~40–90 ns overhead (≈30–36 ns on a real NVMe, ≈86 ns on null-block). Source targets Linux, FreeBSD and SPDK and is slated for Q1 release.

AI-written summary · about

" chapters (25)

1. 0:00Intro
2. 0:44Command Response Protocol
3. 1:05NVME
4. 1:17New Devices
5. 3:51IOU Ring
6. 6:16IOU Ring Limitations
7. 7:22AI Axles
8. 7:45NVME Axles
9. 8:23Linux
10. 8:49PDK
11. 11:03Unified API
12. 13:34Geometry
13. 14:10API
14. 14:43Buff
15. 15:15Buffer
16. 15:30Command Interface
17. 15:48Command Encapsulation
18. 17:31Command Completion
19. 17:58Request Object
20. 18:24Example
21. 20:21Cost
22. 21:37Latency
23. 24:45Nanosecond
24. 25:41Easy API
25. 27:08Command Sets