Read Amplification
The ratio between the amount of data read from storage and the amount of useful data returned to satisfy a logical read request.
Storage Context
Read amplification compares the physical amount of data read with the logical data needed to satisfy a request. An LSM engine may consult several levels or files, while fragmented indexes or fan-out queries can create similar effects at other layers. Bloom filters and compaction can reduce unnecessary reads.
Measurement Boundary
High read amplification does not always translate directly into high latency. Cache hits, parallel I/O, prefetching, and device behavior can change the observed response time.