Optimize Metadata Operations of Key-Value Store via A Flat Indexing LSM-Tree

The effectiveness of applying key-value store mechanisms to manage metadata of file systems has been demonstrated recently. However, traditional indirect metadata indexing schemes are not in concert with key-value data structures, which could degrade the performance of a KV-embedded file system due to the overhead of hierarchical path queries. In this paper, we propose FILT (Flat Indexing LSM-Tree), a lightweight file system middleware that can solve this issue by employing flat indexing. We introduce a new range rename mechanism over LSM-tree to provide fast directory renames. FILT exploits the benefits of both flat indexing and LSM-tree structure to eliminate redundant path look-ups. Our extensive performance evaluation studies show that FILT can offer up to 2.3X performance gain compared with TableFS and 56x to ordinary cloud file systems.