TABLEFS: Enhancing Metadata Efficiency in the Local File System on ATC 2013 by Kai Ren and Garth Gibson.

There are two types of file systems: journaling file system, such as ext4, and copy on write file system, such as btrfs. Both of them use a B-tree like data structure, which is not efficient to manage small metadata intensive workloads because of the write amplification.

This paper propose to use Log-Structured Merge (LSM) tree to improve the metadata update performance by more aggressive aggregation and batching of metadata updates. The prototype directly use an LSM tree implemented by Google’s LevelDB to store the metadata {see: Lock Free LevelDB}. In the evaluation, a write intensive metadata only benchmark is studied. Both the disk read and write request are reduced in TableFS.