6c361c1dfc
Backblaze Drive Stats forbids selling the data itself, which CC BY 4.0 permits, so the failure-rate fields can't be CC BY 4.0. README now splits the licence and a NOTICE documents the Backblaze-sourced fields (cite Backblaze; no resale of the data; no endorsement). drives.json/csv re-exported from the corrected live API. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
70 lines
4.7 KiB
Markdown
70 lines
4.7 KiB
Markdown
# NAS drive dataset
|
|
|
|
An open dataset of NAS hard drives: full specs, **CMR vs SMR**, and real-world **Backblaze annualized failure rates (AFR)**. 138 drives, 13 with a measured failure rate. Our drive specs and CMR/SMR data are free to use - including commercially - under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) with credit; the Backblaze failure-rate fields stay under [Backblaze's own terms](#licence--attribution) (cite Backblaze; don't sell the data itself).
|
|
|
|
Maintained at [nasdisks.com](https://www.nasdisks.com), where the same data drives a live price comparison across seven Amazon regions.
|
|
|
|
## What's inside
|
|
|
|
| File | Rows | What it is |
|
|
|------|------|------------|
|
|
| [`data/drives.csv`](data/drives.csv) | 138 | The headline file: one flat row per drive, specs joined to failure rate. |
|
|
| [`data/drives.json`](data/drives.json) | 138 | The same rows as JSON (mirror of the live API). |
|
|
| [`data/cmr-reference.csv`](data/cmr-reference.csv) | 138 | Source of truth for specs + CMR/SMR classification, with a per-row `confidence` flag. |
|
|
| [`data/backblaze-afr.csv`](data/backblaze-afr.csv) | 36 | Full-year 2025 Backblaze failure rates per model (the models Backblaze runs at scale). |
|
|
| [`data/series-reliability.csv`](data/series-reliability.csv) | - | Manufacturer series specs (warranty, rated workload, MTBF) used as context where Backblaze has no data. |
|
|
|
|
## Quick start
|
|
|
|
The files in `data/` are ready to use as-is. If you'd rather hit a live endpoint (CORS-enabled, cached, no key, no rate limit):
|
|
|
|
```bash
|
|
# JSON: { source, license, attribution, count, drives: [...] }
|
|
curl -s https://www.nasdisks.com/data/drives.json | jq '.drives[0]'
|
|
|
|
# Flat CSV
|
|
curl -s https://www.nasdisks.com/data/drives.csv
|
|
```
|
|
|
|
## Columns (`drives.csv` / `drives.json`)
|
|
|
|
| Column | Meaning |
|
|
|--------|---------|
|
|
| `model` | Manufacturer model number (e.g. `WD80EFPX`). |
|
|
| `brand` | WD, Seagate, Toshiba, HGST. |
|
|
| `line` | Product line (Red Plus, IronWolf Pro, N300, Exos, Ultrastar, etc.). |
|
|
| `capacity_tb` | Capacity in TB. |
|
|
| `rpm` | Rotational speed. |
|
|
| `cache_mb` | DRAM cache in MB. |
|
|
| `interface` | SATA or SAS. |
|
|
| `form_factor` | 3.5 or 2.5. |
|
|
| `recording_tech` | `cmr` or `smr` - the field that matters most for RAID/NAS use. |
|
|
| `drive_class` | NAS, NAS-Pro, Enterprise, Surveillance, Desktop, etc. |
|
|
| `in_production` | Whether the drive is currently sold new. |
|
|
| `afr_pct` | Backblaze annualized failure rate, full-year 2025 (null where Backblaze doesn't track the model). |
|
|
| `reliability_drive_count` / `reliability_drive_days` / `reliability_failures` | The raw figures behind `afr_pct`. |
|
|
| `reliability_source` | Provenance of the failure rate. |
|
|
|
|
## Methodology
|
|
|
|
- **CMR vs SMR** is classified from manufacturer datasheets and model-number decoding. SMR drives shingle their tracks, which cripples random-write and RAID-rebuild performance, so the distinction is the single most important spec for a NAS - and manufacturers do not always advertise it. The `confidence` column in `cmr-reference.csv` flags how certain each call is.
|
|
- **Failure rates** are computed from Backblaze's raw 2025 quarterly Drive Stats, aggregated to a full-year annualized rate and validated against Backblaze's own published fleet figure (~1.36% for 2025). Only the 13 catalog models that exactly match a Backblaze-tracked model carry an `afr_pct`; the rest are left null rather than guessed.
|
|
- **`series-reliability.csv`** gives manufacturer-rated context (warranty years, workload TB/year, MTBF) for lines Backblaze doesn't run, so you have *something* to reason about beyond a blank cell.
|
|
|
|
## Licence & attribution
|
|
|
|
This dataset has two parts under two different sets of terms.
|
|
|
|
**Drive specs + CMR/SMR classification** - everything except the failure-rate fields - is our own work, released under [Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/) (full legal code in [`LICENSE`](LICENSE)). Copy, remix and redistribute it, even commercially, as long as you credit:
|
|
|
|
```
|
|
Drive specs & CMR/SMR by nasdisks.com, CC BY 4.0
|
|
```
|
|
|
|
A link back to https://www.nasdisks.com satisfies it.
|
|
|
|
**Failure-rate fields** - `afr_pct`, `reliability_drive_count`, `reliability_drive_days`, `reliability_failures` (in `drives.csv`/`drives.json`) and the whole of `backblaze-afr.csv` - come from [Backblaze Drive Stats](https://www.backblaze.com/cloud-storage/resources/hard-drive-test-data) and stay under **Backblaze's terms, not CC BY 4.0**: cite Backblaze as the source, and you may not sell the data itself (selling derivative works is fine). Backblaze does not endorse this project. See [`NOTICE`](NOTICE).
|
|
|
|
## Updates
|
|
|
|
The dataset is refreshed as the catalogue and Backblaze data change. The live endpoints above always reflect the current state; this repo is re-exported periodically.
|