Skip to main content

Benchmarking


Operations · Benchmark

Operational objective — Measure the complete workload honestly, including concurrency, dependencies, memory, and tail behavior.
A benchmark that explains saturation and can be repeated across changes.

Key concepts

p50

Typical request

p95

Slow tail

p99

Extreme tail

Evidence model

Evidence over assumption

Do not compare frameworks using different workloads, caches, connection pools, or failure policies.

Operational context

A fast isolated handler can still produce a slow product when database, network, serialization, or queue behavior dominates.
Operator’s principle A benchmark is a decision instrument, not a victory number.

Implementation notes

Decision lens

References

Last modified on September 10, 2026