Write Amplification (WAF): The Hidden Write Tax
Why writing 1 GB to an SSD can cost it 2 GB of real flash writes.
Write amplification is the ratio between the data actually written to the flash and the data the host asked to write. Garbage collection copies valid pages around, so the flash always writes more than the host requested.
WAF = (host writes + GC copy writes) / host writes
A WAF of 1.0 is perfect (no extra writes). A WAF of 3.0 means every 1 GB from the host costs 3 GB of real flash wear.Every GC copy is a write the host never requested — pure overhead that wears the flash and steals bandwidth.
Why you should care
- Endurance — flash cells survive a limited number of erase cycles. Higher WAF burns through that budget faster, shortening drive life.
- Performance — GC copies compete with your real writes for flash bandwidth, causing latency spikes.
- Energy — every extra write costs power.
What makes WAF worse (or better)
- Fuller drive → victims hold more valid data → more copying → higher WAF.
- More over-provisioning → emptier victims → lower WAF.
- Random/skewed writes → mixed hot & cold data in the same block → more copying.
- Smart data placement & GC policy → separates hot from cold → lower WAF.
Analogy. WAF is like moving house every time you buy one new book: you keep re-packing and re-carrying all the books you already owned. The ratio of total books carried to new books bought is your amplification.
In the EyanaSSDSim paper & simulator. Write amplification is the headline metric in EyanaSSDSim. The Live Simulator plots WAF live as a trace replays; the paper compares WAF across workloads, over-provisioning levels, GC policies, and — crucially — between a conventional SSD and ZNS.