perf-benchmarker

Sequential benchmarks prevent flaky results by guaranteeing no parallel runs and enforcing determinism. This approach applies a 10s warmup and enforces a 60s minimum duration per run to ensure reproducible performance measurements.

The minimum duration for a sequential performance benchmark is 60s per run, though binary search benchmarks require 30s. A 10s warmup period is also enforced prior to measurement to stabilize execution.

To establish baselines and validate regressions consistently, run sequential benchmarks with strict duration rules. This prevents parallel interference and outputs a JSON metrics block for reliable cross-run comparisons.

Performance metrics are outputted as a JSON block enclosed between PERF_METRICS_START and PERF_METRICS_END markers. This clear output contract allows automated systems to easily extract and parse the sequential benchmark results.

Yes, the sequential benchmarker supports anomaly re-runs during testing. Combined with enforced minimum durations and warmup periods, this feature helps validate unexpected results and ensures measurement consistency.