Benchmarks
Benchmarks are not publishable until they are reproducible.
Every public run must include provider, region, instance type, date, Valkey version, kernel, dataset size, command mix, concurrency, run id, raw JSON, and Markdown summary. Benchmarks that cannot be regenerated from a checked-in profile are marketing, not measurement — so we gate them until anyone can reproduce the number.
Publication gate
A benchmark is publishable only after every item below is true:
- Provisioned from OpenTofu.
- Configured by Ansible.
- Run from a checked-in profile.
- Backup and restore verified for private-alpha nodes.
- Teardown verified after disposable benchmark runs.
Why this is empty for now
No reproducible run has cleared every gate yet, so there is nothing honest to publish. When the first run does, this page will show the raw JSON, the full environment metadata, and the exact profile used to produce it — not a single headline throughput figure. Until then, the pricing targets elsewhere on the site are labeled as targets, not benchmark results.