ZNS: Avoiding the Block Interface Tax for Flash-based SSDs
summary
ZNS exposes erase-block boundaries and append-only zones so host software moves fine-grained placement into userspace; implemented Linux NVMe+block changes, f2fs zone support, and a RocksDB backend (znsfs). Evaluation uses one 2 TB drive flashed as block (7% and 28% OP) versus ZNS (zero OP) and shows 2–4× lower 99.9th‑percentile random‑read latency for zone‑specialized RocksDB and 2× higher write throughput while maintaining predictable performance and reduced write amplification.
AI-written summary · about
chapters (11)
- 0:00Intro
- 2:15Zoned Namespace SSDs Getting rid of the block interface tax
- 4:23Evolving towards ZNS SSDs
- 5:50Adoption Three ways to adopt ZNS SSDs
- 7:34Enabling the Linux Ecosystem Adding support for ZNS SSDs
- 8:15ZenFS Architecture A new storage backend for RocksDB
- 9:27Evaluation Apples-to-apples comparison
- 10:59Raw 1/0 Characteristics Improving Write Throughput & Read Latency
- 12:15RocksDB: Writes Double the throughput over 28% OP SSDs
- 13:35RocksDB: Reads and Writes Improving Writes and Tail Latencies
- 15:21Summary
related talks
52mHow ZNS SSD's Are Changing the Landscape
27mPreview: Storage
19mA zone-aware cache system for distributed databases
46mZMS a New Lightweight Storage System- 46mZMS a New Lightweight Storage System
30mZiggurat: A Tiered File System for Non-Volatile Main Memories and Disks
40mLinux Storage System Bottleneck for eMMC/UFS
13mLiquid-State Drive: A Case for DNA Block Device for Enormous Data