Benchmarks
Database benchmarks with the raw data attached.
Every run below was measured with YCSB on hardware a customer can buy, and each one has a stable, dated page with its methodology, versions, image digests, hardware spec and caveats. The result tables on those pages are rendered from the CSV and JSON files you can download, so the numbers you cite are the numbers in the files.
All results and raw data are licensed under CC BY 4.0. Reuse them with credit to Layerbase and a link to the run page; each run page has a ready-to-paste citation.
Runs
Measured October 23, 2026
FerretDB vs DocumentDB YCSB benchmark, October 23, 2026YCSB workloads A, B and C against FerretDB 2 and standalone DocumentDB 1.0, each on its own 4 vCPU / 8 GB instance with a 10 million record dataset larger than RAM. Throughput, read p99 and update p99 at 16, 64 and 256 client threads.
Earlier writeups
These runs predate publishing raw files, so they have no run page here. Their measured tables and configuration caveats are in the posts themselves.
Measured August 18, 2026
MongoDB vs FerretDB on the same hardware: the benchmark, and what we had to fix to run it fairlyYCSB against MongoDB 8.0 and FerretDB 2.7 on the same 4 vCPU / 8 GB profile, including the storage and PostgreSQL tuning that changed the result. Measured August 18 and 19, 2026. The results tables are in the post; raw files for this run are not published.
How we publish benchmarks
- A run page is never edited to change its numbers. A re-run gets a new date and a new page.
- Each run states its own caveats, including how many trials it ran and what it does not test.
- Raw log archives stay private when they contain credentials; the published files carry the parsed metrics, pinned versions and the archives' SHA-256 hashes.