HTTP benchmarks (tier5_http)¶

Plan: 2026-05-16-li-httpd-plan.md · Physics bench plan: 2026-05-14-benchmarks-and-simulations.md

Principle¶

Users add scenarios by editing TOML only — not Python. Harnesses merge layers and run verify-before-timing.

TOML layers¶

File	Purpose
`benchmarks/tier5_http/defaults.toml`	Workers, ports, wrk defaults, fixture sizes
`benchmarks/tier5_http/suite.toml`	Profiles (`ci`, `baseline`) and include/exclude
`benchmarks/tier5_http/scenarios/<name>/bench.toml`	Server, verify, load, metrics

Merge order: defaults.toml ← scenarios/*/bench.toml ← --set key=value.

CLI¶

./benchmarks/harness/bench_http.py --profile ci
./benchmarks/harness/verify_http.py --profile ci --stub
./benchmarks/harness/bench_http.py static_small --set load.connections=500

See benchmarks/tier5_http/README.md and params_schema.toml for keys.

Fairness¶

Same URL set, body sizes, worker model, and TLS labels across lang rows. Document variant= when parity is incomplete. CI uses timing = false — no flaky wrk in PR gates.

CSV¶

Same columns as benchmarks/harness/bench.py: benchmark, lang, variant, threads, metric, value, unit, git_sha, cpu_model, flags.

Optional benchmarks/results/verify_http.csv from verify_http.py.

Exploit harness: exploit_http.py + suite_exploits.toml — see security-http-exploits.md.