Benchmarks
Zion's benchmarks live in three places, by altitude:
| Where | What | Output |
|---|---|---|
benchmarks/baseline/ | Canonical reproducible macro-benchmark: throughput (median-of-N trials + 95% CI + p50/p99/p99.9 + CPU/RSS), an nginx comparison on the same box, a concurrency sweep, a payload matrix, HTTP/2 (h2spec) + TLS (testssl) conformance, and cache-correctness — rendered to a tracked PDF. | zion-<ver>-baseline.pdf |
benches/e2e/ | Distributed 2-node rig (load box → SUT over a real NIC) — the network-realistic numbers. | benches/e2e/RESULTS.md |
benches/ | Rust cargo bench microbenches (Criterion): WAF scan, cache lookup, traceparent, audit HMAC, sovereign lookup, NUMA. | criterion baseline |
Specialised one-offs (bench-pgo.sh, bench-mesh.sh, bench-profile.sh) live under benchmarks/.
Honest snapshot
End-to-end TLS 1.3 over loopback to a zero-overhead hyper backend; median of trials; 0 errors / all 2xx. Two reference machines:
- Apple M4 (fast desktop class): TLS reverse proxy ~100–108k req/s (92k for 16 KB bodies), ~101k with the WAF scanning every request, 190–222k for cache hits served from RAM.
- Isolated i7-6700 SUT (3 vCPU,
governor=performance, baseline harness): cache-hit ~55.9k vs nginx ~39.5k req/s on the same box (+41%); proxy passthrough ~27.7k;h2load56k (H2) /wrk59.5k (H1); concurrency-saturation knee ~c=50.
Against nginx 1.27 under identical cgroup limits (1 CPU / 256 MB, byte-identical responses): nginx leads raw uncached proxy by a few %, Zion is at parity on API GET / WAF POST, and wins on cacheable assets where the in-RAM cache skips the origin round-trip — all while terminating TLS and running the WAF inline.
These are server-side-efficiency figures (no network RTT), useful as a cross-version regression baseline — not a WAN number. The full methodology, raw tool output, and the nginx comparison are in the baseline PDF; WAN-realistic numbers are in benches/e2e/RESULTS.md.
Reproduce
# canonical baseline (throughput + conformance + cache-correctness → PDF)
MODE=full bash benchmarks/baseline/run-baseline.sh
# Rust microbenches
cargo bench --no-default-features --bench cache_lookupSee the baseline harness README for prerequisites, knobs (MODE, TRIALS, CPU pinning, REPORT=0), and rigor notes.