Archivestorage
Write Amplification Is the Hidden Tax on Storage
Every byte you write may be rewritten several times before it settles. SSDs charge you for all of it.
- Published
- Reading time
- 1 min read
Write amplification (WA) is bytes physically written divided by bytes logically written. A WA of 10 means a 1 GB ingest writes 10 GB to the device — burning endurance, bandwidth, and compaction CPU.
Where it comes from
- LSM compaction: a key rewritten at each level.
- SSD FTL: garbage collection rewrites whole erase blocks because NAND erases in blocks, not pages.
- RAID: parity updates read-modify-write.
- Filesystem journaling: metadata written twice.
Reducing it
- Tiered or leveled compaction tuned to workload; tiered has lower WA, higher space amp.
- Over-provisioning SSDs gives the FTL room to garbage collect.
- Larger write batches, fewer tiny random writes.
- Compression before writing reduces bytes, but not the number of rewrites.
The trade
Lower WA usually costs read amplification or space amplification. There is no free lunch — pick which one your workload tolerates.
Takeaway: instrument WA per workload, not per database. A configuration that is gentle on ingest can be brutal on range scans.