Explore a conference talk presenting IPLFS, a novel log-structured file system designed to operate without garbage collection. Learn about the two key technical components: IPLFS for infinite partition and Interval Mapping for efficient LBA-to-PBA translation. Discover how separating filesystem partition size from physical storage size and implementing a three-level mapping tree can significantly improve performance. Examine the revised metadata structure, implementation details on Linux kernel 5.11.0, and prototyping on OpenSSD. Gain insights into the performance improvements achieved by eliminating filesystem-level garbage collection, with comparisons to F2FS using FIO and MySQL YCSB A benchmarks. Delve into topics such as discard mechanisms, power failure handling, file system layout, and FTL mapping for infinite partitions.
IPLFS - Log-Structured File System without Garbage Collection
Overview
Syllabus
Intro
Outline
Garbage Collection?
Garbage Collection is costly.
Overhead of Garbage Collection
Why does Garbage Collection exist?
Actively Used File System Partition
Handling very large file system partition
Metadata for Log-structured filesystem
Discard Command
Discard the file system blocks (F2FS)
Discard Mechanism of IPLFS
Discard bitmap with Power failure
Discard Logging Mechanism with Power failure
Recovery of Discard Log
File system Layout of IPLFS
FTL Mapping Table For Infinite Partition?
Moving Mapping Interval
Structure of Interval Mapping
Fixed Region Mapping for Map node
Map Node Compaction
Evaluation
Garbage Collection Elimination.
Conclusion
Taught by
USENIX
