D2FS: Device-Driven Filesystem Garbage Collection
summary
Claims that device-side GC can replace filesystem GC and cut overhead by letting the SSD reclaim invalid segments and push LBA remaps to the host. Implements three components: Coupled Garbage Collection, Migration Upcall, and Virtual Overprovisioning (2.4× virtual FS partition, regular region 1.4×, GC region = device capacity). Uses an emulated 256 GB SSD with 32 MB GC unit, FIO and MySQL YCSB-F; D2FS removes costly FS GC (FS GC is ~3× slower) and yields ~3× FIO, 1.7× zoned F2FS, 1.5× IPLFS gains.
AI-written summary · about
chapters (11)
- 0:00D2FS introduction and motivation
- 1:20Redundant GC problems
- 3:26Device-driven GC key idea
- 5:11Design challenges in device-driven GC
- 6:55D2FS architecture overview
- 7:37Coupled GC mechanism
- 8:42Migration UPC for host-device sync
- 10:06Virtual over-provisioning in D2FS
- 12:10Evaluation setup and performance
- 13:58Summary of findings
- 14:21Q&A
related talks
19mIPLFS: Log-Structured File System without Garbage Collection
24mSSDFS: ZNS SSD ready file system with zero GC overhead
36mFlash Friendly File System
23mImproving data placement for Zoned Linux File systems
17mDSA-2LM: A CPU-Free Tiered Memory Architecture with Intel DSA
21mThe Design and Implementation of a Capacity-Variant Storage System
31mDesign and implementation of Autocaching for CXLSSD
14mUnleashing Zoned UFS: Cross-Layer Optimizations for Next-Generation Mobile Storage