Tiered Compaction

Tiered compaction groups similarly sized sorted components and merges several of them together when a policy threshold is reached. Compared with a heavily leveled arrangement, it can reduce how often new data is rewritten.

The tradeoff is that reads may search more components and obsolete versions may occupy more space before merging. Implementations use different names and details, so the relevant engine’s policy should be checked before applying a tuning assumption.