Improving data placement for Zoned Linux File systems
summary
Builds a zoned-device-aware file layout (zenfs-style) that separates files by lifetime and writes to zone-aligned regions, integrates with RocksDB integration and Percona MySQL packages, and benchmarks against XFS on the same media. Results show better throughput and QoS for heavy write and mixed workloads when drives are prefilled to 90% full to trigger GC, largely by avoiding device-side GC at the cost of ~5% space amplification. Works well when file sizes map to zones or when lifetime hints are available; tiny files still require compaction and some host GC.
AI-written summary · about
chapters (7)
- 0:00Introduction and motivation
- 3:12Problem definition and ZNS benefits
- 8:55Use cases and performance results
- 16:05Lessons for general file systems
- 19:17Directory-based placement hints
- 21:02Garbage collection and tail latency
- 23:00Q&A and summary
related talks
25mUsing Zoned Block Devices in Fedora
14mZNS+: Advanced Zoned Namespace Interface for Supporting In-Storage Zone Compaction
31mFrom Open-Channel SSDs to Zoned Namespaces
14mUnleashing Zoned UFS: Cross-Layer Optimizations for Next-Generation Mobile Storage
45mData Placement in Linux: Evolving Block and File I/O
15mD2FS: Device-Driven Filesystem Garbage Collection
30mOpenZFS and Linux
38mRethinking Writeback: Scaling Linux Filesystem and Memory Performance for the Next Decade