lazy-leveling · research

Lazy leveling

Lazy leveling combines tiering at smaller LSM-tree levels with leveling at the largest level.

Dayan and Idreos introduce this policy in Dostoevsky. Its point-lookup analysis co-optimizes Bloom-filter allocation; short range lookups retain a larger startup cost. It still performs merges when data moves down the tree. [1]

Relations

related

Fluid LSM-tree, Leveled compaction, Tiered compaction, Sorted run, Bloom filter

Citations

1. Dayan and Idreos. Dostoevsky. SIGMOD 2018, pp. 505-520; sections 4.1-4.3 and Figure 8. DOI: 10.1145/3183713.3196927. (accessed 2026-11-10)

JSON: /data/terms/lazy-leveling.json