fast-telemetry benchmark overview
This page ties together the benchmark data referenced in the launch post. Use it as the hub for the full benchmark story, then drill into the dedicated Criterion and threaded harness pages for deeper detail.
What to open first
Use this when you want isolated benchmark families: recording hot paths, export cost, dynamic-series behavior, and span serialization work.
Use this when you want the contention-heavy story: labeled and dynamic metric paths, span scenarios, and the biggest speedups from the local run.
Reading notes
- The threaded harness report below comes from one local run and should be treated as workload-specific, not universal.
- Criterion microbenchmarks cover recording hot paths, exporter cost, dynamic-series first touch, write-plus-export overlap, and span OTLP pipeline cost.
- The harness focuses on sustained multi-threaded contention scenarios, which is where sharded write paths matter most.
- Atomic baseline numbers only appear for the plain counter cases in the harness report.
Representative results
The headline data points from the launch post, taken directly from the harness run.
| Harness case | Speedup | fast ops/sec | OTel ops/sec |
|---|---|---|---|
| labeled_counter:churn | 2,364x | 8.8B ops/sec | 3.7M ops/sec |
| dynamic_counter:uniform | 1,743x | 6.1B ops/sec | 3.5M ops/sec |
| pipeline | 324x | 59.4M ops/sec | 183.1K ops/sec |
| root | 253x | 267.3M ops/sec | 1.1M ops/sec |
| dynamic_gauge_i64:uniform | 9x | 32.2M ops/sec | 3.6M ops/sec |
Criterion
Criterion is the unit benchmark suite for specific features, so it is the best place to inspect isolated recording, export, and span paths.
Harness
The threaded harness is a full lifecycle test under simulated CPU load, so it is the better view for sustained contention and end-to-end throughput behavior.