Validation
Validation — EyanaSSDSim vs FEMU (WAF)
What it shows. Write Amplification Factor (WAF = physical writes / host writes) from EyanaSSDSim placed next to FEMU (a hardware-accurate NVMe emulator) for eight configurations measured at the same over-provisioning (OP 25%).
How FEMU was measured. Conventional points use fio random-write
fill (OP set by fill size, 3× rewrite for steady state); FDP points use a custom
fdp_write tool that steers each 4 KB write to a placement handle via
NVMe pass-through. WAF is read from FEMU's instrumented FTL counters
(host/gc pages, hbmw/mbmw).
The result — two regimes:
- Synthetic (uniform, 80/20): ±3–8%, mixed signs. Errors cancel; the simulator tracks FEMU with no systematic bias where GC is mild and uniform.
- Real traces (prxy): −18% to −29%, one-directional. The simulator is consistently below FEMU. This is not error — it is the garbage-collection erase granularity: the simulator's default block-level reclaim migrates fewer valid pages than FEMU's superblock (line-level) erase under the heavy, skewed GC that real traces induce.
Takeaway. The gap is bounded, explained, one-directional, and closes when the simulator is configured for superblock GC (see the Granularity Bracket graph).
What we found
- Synthetic, sim vs FEMU: uniform conv +7.1% (1.56 vs 1.46), uniform FDP-single −7.5% (1.54 vs 1.66), uniform FDP-PI +6.1% (1.75 vs 1.65), 80/20 FDP-single +3.2% (1.16 vs 1.12), 80/20 FDP-PI +8.0% (1.10 vs 1.02) — errors cancel, no bias.
- Real traces, sim vs FEMU: prxy_0 conv −17.8% (1.54 vs 1.87), prxy_1_3M FDP-single −19.6% (1.26 vs 1.56), FDP-PI −28.8% (1.06 vs 1.49) — systematically low, one direction.
- Trend fidelity: the ordering of configs is preserved (Spearman ρ = 1.0); the simulator is always ≤ FEMU on real traces, exactly as the granularity model predicts.
