YAPP · yet another pool planner
Find the perfect pool
Pick what you'll run on it. We'll search the catalog and recommend the cheapest pool that comfortably sustains the workload.
Or start from a preset
Describe your pool
Edit any chip below — the blueprint and stats update live.
Live blueprint
— spec a pool, see the splash —
Load simulation
First-order steady-state estimate · pick a workload, see if the pool sustains it · not a substitute for fio
Saved pools
Drive catalog
Real, currently-shipping drives — specs from manufacturer datasheets · live price/TB per region
Drive Tier Cap Price/TB MTBF (M h) AFR URE Ref
▾ How these numbers are sourced — direct vs. derived

When you pick a drive, its MTBF, URE rate, and per-disk IOPS/throughput are used directly in the failure & rebuild model — replacing the generic "HDD / SSD / NVMe" defaults. Per-disk throughput is capped at 4× the type baseline so a single fast NVMe drive doesn't project unrealistic pool-level numbers (controllers, CPU, and ZFS itself become the bottleneck long before raw drive throughput does).

What the datasheets actually publish:

  • MTBF / URE — published for almost all enterprise & NAS drives. Most consumer HDD datasheets (BarraCuda, WD Blue, Toshiba P300) and Samsung consumer SSDs omit URE; the catalog uses conservative class estimates and labels them as such in source comments.
  • SSD / NVMe random IOPS & sequential MB/s — published in every datasheet we cite. Catalog values are taken verbatim for the representative capacity (largest typically; 1 TB SKUs cap lower).
  • HDD sequential MB/s — published; capacity-dependent. Catalog uses the most representative capacity per family.
  • HDD random IOPS — Seagate Exos and WD Ultrastar publish them (4 K QD16 / QD32); catalog uses datasheet figures. NAS, surveillance, and consumer HDDs do not publish random IOPS. For those models the catalog uses ~120 IOPS for 7200 rpm and ~80–100 IOPS for 5400 rpm, derived from average rotational latency + seek time. Treat HDD random IOPS as physics estimates, not vendor numbers.
  • Price/TB — fetched live from every retailer for the selected region (cheapest wins), fallback baked-in USD estimate. Always an estimate, never a quote.
Failure model & references ▾
Disk failure rate (MTBF / AFR)

Per-disk failure follows an exponential model with rate λ = 1 / MTBF. Annual Failure Rate ≈ 8760 / MTBF. Defaults used when you select "Custom drive":

TypeMTBF (h)AFRSource
HDD 1,200,000~0.73 %Seagate Enterprise spec [A1]
SSD 2,000,000~0.44 %Enterprise SSD datasheet typical
NVMe 2,500,000~0.35 %Datacenter NVMe class [C4]

Real-world fleet data tracks somewhat higher: Backblaze reported a 1.36 % annual failure rate across 344k drives in 2025 (down from 1.55 % in 2024). Treat the spec MTBF as a best case. [D1]

Unrecoverable Read Error (URE) risk

Probability of at least one URE while reading B bits during resilver: P = 1 − e−B·r, where r is the per-bit error rate.

ClassURE rate≈ 1 error per
Consumer HDD (e.g. WD Red Plus)1 in 101412.5 TB read
Enterprise / NAS HDD (e.g. Exos, IronWolf Pro)1 in 1015125 TB read
Datacenter SSD1 in 10161.25 PB read
NVMe (consumer / DC)1 in 101712.5 PB read

Manufacturer datasheets quote URE as a worst-case bound, not an average — measured rates are often 10–20× lower in the field. [D2]

Resilver-window data loss

After one disk fails, the vdev can survive up to tol more failures (mirror = N−1, RAID-Zn = n). Probability of exceeding tolerance during resilver window t is approximated by a binomial tail with p = 1 − e−λt per surviving disk. URE risk during the read-every-bit rebuild is added on top.

MTTDL (mean time to data loss) ≈ 1 / (AFRpool · Ploss|failure). This is a model, not a guarantee — correlated failures (same batch, shared PSU, vibration) and silent corruption aren't captured. Always have backups.

References

Drive datasheets

  1. [A1] Seagate Exos X18 / X20 / X22 product page — 2.5 M h MTBF, 1 sector per 1015 bits, 285 MB/s sustained.
  2. [A2] WD Gold Enterprise HDD datasheet — 2.5 M h MTBF, 1 in 1015 URE.
  3. [A3] Toshiba MG10 series datasheet — 2.5 M h MTBF, 1 in 1015 URE.
  4. [A4] Seagate IronWolf Pro / IronWolf product page — 1.2 M h MTBF (Pro) / 1 M h MTBF, 300 TB/yr workload rating.
  5. [A5] WD Red Pro / Red Plus / Purple datasheet — 1 M h MTBF, 1 in 1014 URE.
  6. [A6] Seagate BarraCuda · WD Blue · Toshiba P300 — desktop class, 600 k–1 M h MTBF, 1 in 1014 URE.
  7. [A7] WD Ultrastar DC HC550 · HC560 · HC570 · HC580 datasheets — 7200 rpm helium, 2.5 M h MTBF, 1 in 1015 URE, 269–298 MB/s sustained.
  8. [B1] Samsung 870 EVO product brief — 1.5 M h MTBF, 560/530 MB/s seq, 98k/88k IOPS.
  9. [B2] Samsung 870 QVO product brief — QLC NAND, 1.5 M h MTBF.
  10. [B3] Crucial MX500 product brief — 1.8 M h MTBF, 95k/90k IOPS.
  11. [B4] Samsung PM893 datacenter SATA datasheet — 2 M h MTBF, 97k/31k IOPS.
  12. [C1] Samsung 990 Pro datasheet — PCIe 4.0, 1.55M/1.4M IOPS, 7.45 / 6.9 GB/s.
  13. [C2] WD Black SN850X datasheet — 1.75 M h MTBF, 1.2M/1.1M IOPS.
  14. [C3] Samsung PM9A3 U.2 datasheet — datacenter NVMe, 2 M h MTBF.
  15. [C4] Solidigm D7-P5520 datasheet — 2.5 M h MTBF, 1 sector per 1017 bits.
  16. [C5] Kioxia CD8-R datasheet — datacenter NVMe, 2.5 M h MTBF.

Reliability research

  1. [D1] Backblaze Drive Stats 2024 — 1.36 % fleet AFR across 344,196 drives.
  2. [D2] Bairavasundaram et al., An Analysis of Data Corruption in the Storage Stack (FAST '08) — observed unrecoverable read error rates 1–2 orders of magnitude lower than spec.

All probabilities are first-order estimates suitable for comparing layouts. They aren't a substitute for proper reliability modelling on a specific hardware fleet.