Benchmark
Numbers from labyrinth-bench on a single AMD Ryzen 9 9950X3D thread — and
the tool to reproduce them on your own hardware.
Results
| Operation | ops/sec | Latency |
|---|---|---|
| Cache Get | 22M | 45 ns |
| Cache Set | 19M | 53 ns |
| Wire Unpack | 4.4M | 225 ns |
| Wire Pack | 2.6M | 391 ns |
| Name Decode | 22.6M | 44 ns |
| FNV-1a Hash | 331M | 3 ns |
| Full Resolve (cached) | 700K | 1.4 µs |
Binary size: 6.8 MB stripped, with the web dashboard embedded.
What is measured
Single-thread microbenchmarks of the hot paths: the 256-shard cache, wire format packing and unpacking, name decoding, hashing, and an end-to-end cached resolve. The same code paths serve production traffic.
Run it yourself
labyrinth-bench binaries ship with every release — Linux, macOS and
Windows, amd64 and arm64, plus checksums:
- Download the
labyrinth-bench-<os>-<arch>asset from the releases page. - Verify it against
checksums.txt. - Run it on the hardware you plan to deploy.
Throughput numbers scale with your hardware; the ratios between operations are what to compare.
Source: the project README. See also: Caching, Performance tuning.