1st Storage System, Hard Disk and Solid State Technologies Summit, IEEE Asia-Pacific Magnetic Recording Conference (APMRC), November 2012, Singapore.
Kai Ren and Garth A. Gibson
Carnegie Mellon University
Pittsburgh, PA 15213
Conventional file systems are optimzed for large file transfers instead of workloads that are dominated by metadata and small file accesses. This paper examines using techniques adopted from NoSQL databases to manage file system metadata and small files, which feature high rates of change and efficient out-of-core data representation. A FUSE file system prototype was built by storing file system metadata and small files into a modern key-value store: LevelDB. We demonstrate that such techniques can improve the performance of modern local file systems in Linux for workloads dominated by metadata and tiny files.
KEYWORDS: System software, File systems, Metadata representation, Log-structure Merge Tree.
FULL PAPER: pdf