This suite measures the raw HTTP-to-Fetch adapter cost of the published and development versions of @hono/node-server. A native node:http implementation is the performance floor, while srvx with FastResponse is another Fetch-compatible adapter for comparison.

The benchmark intentionally avoids Hono routing so application-framework work does not obscure adapter work.

• Node.js
• oha

pnpm install
pnpm run -w build
pnpm run benchmark

To replace the results below after a full successful run:

pnpm run benchmark --update

Quick smoke run:

BENCH_CONNECTIONS=10 BENCH_WARMUP=1s BENCH_DURATION=1s BENCH_TRIES=1 pnpm run benchmark

The defaults are 100 concurrent connections, a discarded two-second warmup, and the median of three five-second measurements. The server order is randomized to reduce systematic thermal or background-load bias.

The suite validates every response before measuring it, rejects load-generator errors or unexpected status classes, and reports request rate, mean latency, response throughput, and isolated server peak RSS.

It covers:

1. A bodyless 204 response to HEAD
2. A small text response
3. URL and query-string parsing
4. Incoming header reads and outgoing headers
5. JSON serialization
6. JSON request parsing and response serialization
7. A 64 KiB request body
8. A fixed 64 KiB response body
9. A chunked ReadableStream response

These are synthetic microbenchmarks. They are useful for finding adapter regressions, not for predicting application throughput in production.

The benchmark methodology and reporting are inspired by the srvx Node.js compatibility benchmarks. The endpoint suite builds on ideas from bun-http-framework-benchmark.

CPU:        13th Gen Intel(R) Core(TM) i5-13450HX
Node.js:    v24.19.0
OS:         linux x64
OHA:        oha 1.15.0
Config:     100 connections, 2s warmup, 3 × 5s