v1.0.3 - refresh Backblaze failure rates to a rolling window through Q1 2026
This commit is contained in:
parent
d4b82c5de0
commit
f4ab5a584f
10 changed files with 319 additions and 204 deletions
25
README.md
25
README.md
|
|
@ -4,7 +4,7 @@
|
|||
|
||||
**Cite:** NASdisks (2026), *NAS Drive Dataset*, Zenodo. doi:[10.5281/zenodo.21225049](https://doi.org/10.5281/zenodo.21225049)
|
||||
|
||||
An original dataset of NAS and enterprise hard drives: full specs, **CMR vs SMR classification**, and **annualized failure rates** derived from Backblaze Drive Stats. 149 drives.
|
||||
An original dataset of NAS and enterprise hard drives: full specs, **CMR vs SMR classification**, and **annualized failure rates** derived from Backblaze Drive Stats. 208 drives.
|
||||
|
||||
The core of this dataset - the CMR/SMR classification and drive specs - is our own original research, compiled from manufacturer datasheets and model-number analysis. Manufacturers do not publish a unified CMR/SMR list; we built one. The failure-rate column enriches it with rates derived from Backblaze's raw Drive Stats data (our computation, their underlying counts).
|
||||
|
||||
|
|
@ -16,11 +16,11 @@ Maintained at [nasdisks.com](https://www.nasdisks.com/), where the same data pow
|
|||
|
||||
| File | Rows | What it is |
|
||||
|------|------|------------|
|
||||
| [`data/drives-2026-06-26.csv`](data/drives-2026-06-26.csv) | 149 | The headline file: one flat row per drive, all specs and failure rate combined. |
|
||||
| [`data/drives-2026-06-26.json`](data/drives-2026-06-26.json) | 149 | The same rows as JSON (mirror of the live API). |
|
||||
| [`data/cmr-reference-2026-06-26.csv`](data/cmr-reference-2026-06-26.csv) | 149 | Source of truth for specs and CMR/SMR classification (including idle/seek noise and helium fill), with a per-row `confidence` flag. Our original research. |
|
||||
| [`data/backblaze-afr-2026-06-26.csv`](data/backblaze-afr-2026-06-26.csv) | 36 | Full-year 2025 annualized failure rates, computed from Backblaze raw Drive Stats. |
|
||||
| [`data/series-reliability-2026-06-26.csv`](data/series-reliability-2026-06-26.csv) | - | Manufacturer series specs (warranty, rated workload, MTBF) used as context where Backblaze has no data. |
|
||||
| [`data/drives-2026-07-16.csv`](data/drives-2026-07-16.csv) | 208 | The headline file: one flat row per drive, all specs and failure rate combined. |
|
||||
| [`data/drives-2026-07-16.json`](data/drives-2026-07-16.json) | 208 | The same rows as JSON (mirror of the live API). |
|
||||
| [`data/cmr-reference-2026-07-16.csv`](data/cmr-reference-2026-07-16.csv) | 208 | Source of truth for specs and CMR/SMR classification (including idle/seek noise and helium fill), with a per-row `confidence` flag. Our original research. |
|
||||
| [`data/backblaze-afr-2026-07-16.csv`](data/backblaze-afr-2026-07-16.csv) | 31 | Annualized failure rates over a rolling window through Q1 2026 (full-year 2025 plus the Q1 2026 quarter), computed from Backblaze raw Drive Stats. |
|
||||
| [`data/series-reliability-2026-07-16.csv`](data/series-reliability-2026-07-16.csv) | - | Manufacturer series specs (warranty, rated workload, MTBF) used as context where Backblaze has no data. |
|
||||
|
||||
## Quick start
|
||||
|
||||
|
|
@ -34,7 +34,7 @@ curl -s https://www.nasdisks.com/data/drives.json | jq '.drives[0]'
|
|||
curl -s https://www.nasdisks.com/data/drives.csv
|
||||
```
|
||||
|
||||
## Columns (`drives-2026-06-26.csv` / `drives-2026-06-26.json`)
|
||||
## Columns (`drives-2026-07-16.csv` / `drives-2026-07-16.json`)
|
||||
|
||||
| Column | Meaning |
|
||||
|--------|---------|
|
||||
|
|
@ -52,16 +52,16 @@ curl -s https://www.nasdisks.com/data/drives.csv
|
|||
| `drive_class` | NAS, NAS-Pro, Enterprise, Surveillance, Desktop, etc. |
|
||||
| `in_production` | Whether the drive is currently sold new. |
|
||||
| `also_sold_as` | Equivalent SKUs merged into this drive - format/firmware/secure-erase/OEM variants of the same physical drive. `;`-separated in the CSV, an array in JSON; empty when none. |
|
||||
| `afr_pct` | Annualized failure rate, full-year 2025, derived from Backblaze Drive Stats (null where Backblaze doesn't track the model). |
|
||||
| `afr_pct` | Annualized failure rate over a rolling window through Q1 2026 (2025 + the Q1 2026 quarter), derived from Backblaze Drive Stats (null where Backblaze doesn't track the model). |
|
||||
| `reliability_drive_count` / `reliability_drive_days` / `reliability_failures` | The raw Backblaze figures behind `afr_pct`. |
|
||||
| `reliability_source` | Provenance of the failure rate. |
|
||||
|
||||
## Methodology
|
||||
|
||||
- **CMR vs SMR** is our original classification, built from manufacturer datasheets and model-number decoding. Manufacturers ship SMR drives into NAS lines without labelling them as such; this dataset exists because there was no single authoritative per-model reference. The `confidence` column in `cmr-reference-2026-06-26.csv` flags certainty per row.
|
||||
- **CMR vs SMR** is our original classification, built from manufacturer datasheets and model-number decoding. Manufacturers ship SMR drives into NAS lines without labelling them as such; this dataset exists because there was no single authoritative per-model reference. The `confidence` column in `cmr-reference-2026-07-16.csv` flags certainty per row.
|
||||
- **Noise and helium** are read from manufacturer datasheets (idle/seek A-weighted sound power; helium-seal designation). Our compilation; licensed CC BY 4.0 alongside the other specs.
|
||||
- **Failure rates** are computed from Backblaze's raw 2025 quarterly Drive Stats: we aggregate drive-days and failure counts across quarters into a single full-year annualized rate and validate against Backblaze's published fleet figure (~1.36% for 2025). The computation and methodology are ours; the underlying counts are Backblaze's (see licence section). Models Backblaze doesn't track are left null rather than guessed.
|
||||
- **`series-reliability-2026-06-26.csv`** gives manufacturer-rated context (warranty, workload TB/year, MTBF) for lines Backblaze doesn't run.
|
||||
- **Failure rates** are computed from Backblaze's raw quarterly Drive Stats: we aggregate drive-days and failure counts into a single annualized rate over a rolling window through Q1 2026 (full-year 2025 plus the Q1 2026 quarter), which keeps per-model samples large, and validate against Backblaze's published fleet figures (~1.34% over that window). The computation and methodology are ours; the underlying counts are Backblaze's (see licence section). Models Backblaze doesn't track are left null rather than guessed.
|
||||
- **`series-reliability-2026-07-16.csv`** gives manufacturer-rated context (warranty, workload TB/year, MTBF) for lines Backblaze doesn't run.
|
||||
|
||||
## Licence & attribution
|
||||
|
||||
|
|
@ -80,7 +80,7 @@ Our original work (every field *except* the failure-rate ones). Licensed under [
|
|||
|
||||
### 2. Failure-rate data - Backblaze's terms (NOT CC BY 4.0)
|
||||
|
||||
The fields `afr_pct`, `reliability_drive_count`, `reliability_drive_days`, `reliability_failures` (in `drives-2026-06-26.csv` / `drives-2026-06-26.json`) and all of `backblaze-afr-2026-06-26.csv` are derived from Backblaze Drive Stats. The computation is ours; the underlying counts are Backblaze's.
|
||||
The fields `afr_pct`, `reliability_drive_count`, `reliability_drive_days`, `reliability_failures` (in `drives-2026-07-16.csv` / `drives-2026-07-16.json`) and all of `backblaze-afr-2026-07-16.csv` are derived from Backblaze Drive Stats. The computation is ours; the underlying counts are Backblaze's.
|
||||
|
||||
- **You can:** use it for free, and build (and even sell) derivative works from it.
|
||||
- **You must:** credit Backblaze as the source.
|
||||
|
|
@ -93,6 +93,7 @@ This repo is a point-in-time snapshot; the date in each filename is its export d
|
|||
|
||||
## Changelog
|
||||
|
||||
- **2026-07-16 - v1.0.3** - Refreshed Backblaze failure rates to a rolling window through Q1 2026 (full-year 2025 plus the Q1 2026 quarter, ~1.34% fleet); 208 drives, 25 with a failure rate.
|
||||
- **2026-06-26 - v1.0.2** - Refreshed to the current live catalog (149 drives); minor catalog reconciliation since v1.0.1.
|
||||
- **2026-06-21 - v1.0.1** - Added per-drive noise (idle/seek A-weighted sound power, dB) and helium fields. Corrected WD40EZAX / WD60EZAX to CMR (newer EZAX is CMR; only older EZAZ is SMR).
|
||||
- **2026-06-16 - v1.0.0** - Initial public dataset: full specs, CMR/SMR classification, full-year 2025 Backblaze annualized failure rates.
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue