One write becomes many
Write amplification is the ratio of physical writes performed by the storage layer to the logical writes requested by the application. When you save one record, the system may actually write that data several times in different places.
Where it comes from
- Log structured storage like an LSM tree first writes to a log, then to memory, then flushes to disk, then compacts by rewriting data into merged files. Each level rewrite is extra physical writing.
- Solid state drives erase and rewrite whole blocks even for a small change, amplifying at the hardware level.
- Indexes and replicas each multiply the write further.
Why it matters
Amplification consumes disk bandwidth and, on flash, wears out the device faster since each cell tolerates a limited number of writes. A high ratio can throttle throughput and shorten hardware life.
Tuning compaction strategy trades write amplification against read amplification and space amplification, the classic three way storage tradeoff.
Key idea
Write amplification is the multiplier between logical and physical writes, driven by compaction, flash erase blocks, indexes, and replicas.