Dostoevsky slashes LSM-tree merge overhead by merging only the largest level
Better Space-Time Trade-Offs for LSM-Tree Based Key-Value Stores [pdf]
Dostoevsky introduces Lazy Leveling, a design that eliminates merge operations from all but the largest level of an LSM-tree, improving update cost without degrading point lookups, long range lookups, or space amplification. The Fluid LSM-tree generalizes the design space, and Dostoevsky adaptively navigates it based on workload and hardware. Implemented on RocksDB, it strictly dominates state-of-the-art key-value stores in performance and storage space.
Since Dostoevsky spans all existing designs and is able to navigate to the best one for a given application, it strictly dominates existing key-value stores in terms of performance and space-amplification.