Reducing garbage collection overhead of log-structured file systems with GC journaling

The log-structured file system (LFS) writes all modifications to storage sequentially with append-only logging scheme. This characteristic of LFS is very advantageous to flash storages since the flash memory does not permit in-place overwrite. However, LFS has a high garbage collection (GC) overhead. In particular, under the lazy metadata update scheme, each GC process should invoke the high-cost checkpointing which flushes all the dirty metadata and normal data to storage. The long GC latency will degrade the response times of user requests. In this paper, we propose a GC journaling technique, which journals only the file system changes relevant to the GC process without invoking the high cost checkpointings.