Performance

Sova vs Axum vs Actix-web on an identical multi-page fixture site (same HTML/JSON bodies, verified SHA-256).

Methodology

• Stand: bench/stand/ — shared fixtures in fixtures/, three minimal servers (stand_sova, stand_axum, stand_actix).
• Bodies must match byte-for-byte across frameworks before load runs (run.sh aborts on mismatch).
• Load tool: oha.
• This capture: duration 10s, concurrency 50, TOKIO_WORKER_THREADS=4.
• Captured at 2026-08-08T09:42:22Z on Admins-MacBook-Pro.local.

Pages: /, /about, /blog, /blog/hello, /contact, /api/health.

Latest results — GET /

Framework	Req/s	p50 (ms)	p99 (ms)
sova	101685	0.415	1.725
axum	120476	0.403	0.820
actix	85351	0.509	2.347

Latest results — mean across all paths

Framework	Mean Req/s	Mean p99 (ms)
sova	103094	1.705
axum	117216	1.006
actix	92173	1.899

Per-path detail

Framework	Path	Req/s	p50 (ms)	p99 (ms)
sova	/	101685	0.415	1.725
axum	/	120476	0.403	0.820
actix	/	85351	0.509	2.347
sova	/about	112790	0.411	1.181
axum	/about	118332	0.405	0.880
actix	/about	94307	0.510	1.659
sova	/blog	100645	0.414	1.671
axum	/blog	116968	0.404	1.046
actix	/blog	91560	0.504	2.260
sova	/blog/hello	103288	0.412	1.624
axum	/blog/hello	117281	0.408	0.944
actix	/blog/hello	94716	0.512	1.631
sova	/contact	96406	0.418	2.243
axum	/contact	113758	0.407	1.146
actix	/contact	93758	0.518	1.593
sova	/api/health	103748	0.408	1.784
axum	/api/health	116484	0.398	1.198
actix	/api/health	93347	0.507	1.902

Re-run / regression gate

./bench/stand/run.sh
./bench/stand/run.sh --update-baseline   # after intentional perf changes
DURATION=15s CONCURRENCY=100 ./bench/stand/run.sh

Regression thresholds (vs bench/stand/results/baseline.json): home RPS drop > 15% or p99 rise > 40% fails the script.

Machine-sensitive: compare relative rankings and deltas, not absolute RPS across laptops.