IPLFS: Log-Structured File System without Garbage Collection
summary
IPLFS implements a log‑structured filesystem that stops doing filesystem‑level garbage collection by setting a very large infinite logical partition (64‑bit sector space) and sending aggressive discards. It replaces F2FS’s block bitmap with an in‑memory discard bitmap plus on‑disk discard logging and ships an Interval Mapping three‑level FTL (root/zone/map nodes) prototype on OpenSSD and Linux 5.11.0. Results show IPLFS eliminates FS GC, but still faces device GC, and outperforms F2FS by up to 12.8× (FIO) and 3.73× (MySQL YCSB A).
AI-written summary · about
chapters (23)
- 0:00Intro
- 0:21Outline
- 2:07Garbage Collection?
- 2:29Garbage Collection is costly.
- 2:50Overhead of Garbage Collection
- 3:08Why does Garbage Collection exist?
- 4:41Actively Used File System Partition
- 5:40Handling very large file system partition
- 6:50Metadata for Log-structured filesystem
- 7:30Discard Command
- 7:53Discard the file system blocks (F2FS)
- 9:33Discard Mechanism of IPLFS
- 9:59Discard bitmap with Power failure
- 11:16Discard Logging Mechanism with Power failure
- 11:48File system Layout of IPLFS
- 12:12FTL Mapping Table For Infinite Partition?
- 13:15Moving Mapping Interval
- 13:36Structure of Interval Mapping
- 14:46Fixed Region Mapping for Map node
- 15:23Map Node Compaction
- 17:06Evaluation
- 17:25Garbage Collection Elimination.
- 18:29Conclusion
related talks
15mD2FS: Device-Driven Filesystem Garbage Collection
36mFlash Friendly File System
19mLLFree: Scalable and Optionally-Persistent Page-Frame Allocation
45mData Placement in Linux: Evolving Block and File I/O
23mImproving data placement for Zoned Linux File systems
12mRethinking File Mapping for Persistent Memory
12mpFSCK: Accelerating File System Checking and Repair for Modern Storage
21mThe Design and Implementation of a Capacity-Variant Storage System