HTTP Server Benchmarks
wrk — 120s per test — Linux 7.2.6 (AMD Ryzen 9 7950X) — taskset CPU pinning
httpcats 0.3.2 · miou 0.10.1 · vif 0.0.1~beta5 · httpun 0.2.0 + eio 1.6 · nginx 1.30.5 · OCaml 5.5.0
Scalability: Req/s by number of domains
12 clients / 12 threads. Shows how each implementation leverages multi-domain parallelism.
Requests per second — 8 domains (average over 3 attempts)
Higher is better. The percentage on each bar indicates the wrk ±σ.
Average latency — 8 domains (µs)
Lower is better. The percentage on each bar indicates the wrk ±σ.
Latency vs Throughput (scatter plot — all runs)
Each dot = 1 wrk run. The ideal corner is bottom-right. Dot size reflects the number of domains.
Stability: wrk ±σ by server and configuration
Heatmap of the percentage of requests within ±1 standard deviation (latency). Greener is more stable.