Skip to content

benchmarking.md

Latest commit

 

History

History
53 lines (33 loc) · 2.27 KB

benchmarking.md

File metadata and controls

53 lines (33 loc) · 2.27 KB
Raw

Benchmarking

Benchmarks live in bench/, its own package (@panmdaa/server-bench) with a devDependency on ws, hono, find-my-way, express, fastify, elysia, autocannon, tsx.

How to run

npm run bench                # all benches
npm run bench:router         # RadixTree vs hono vs find-my-way
npm run bench:http           # HTTP server vs node/express/fastify/hono/elysia
npm run bench:http:post      # JSON POST bodies, same competitors
npm run bench:middleware     # compileHandler chains vs direct call
npm run bench:parse          # parseQuery/parseCookies vs baselines
npm run bench:ws             # pipelined echo clients — panmdaa vs ws baseline

Tuning knobs via env vars (WS bench): BENCH_DURATION, BENCH_CONNECTIONS, BENCH_PIPELINE, BENCH_PAYLOAD, BENCH_WARMUP.

What each bench measures

Router (bench/src/router.bench.ts)

RadixTree vs hono's RegExpRouter vs find-my-way, across the route tables in shared.ts:

  • ROUTES — realistic mixed static/dynamic/wildcard.
  • DEEP_ROUTES — deeply nested dynamic routes.
  • LARGE_ROUTES — 1000 generated routes for scalability.

HTTP (bench/src/http-bench.ts, http-post-bench.ts)

autocannon against node, panmdaa, express, fastify, hono, elysia — static and dynamic GET, plus JSON POST. Each server is warmed up before measuring (the first server measured runs cold and the last one hot, http-bench.ts:74-82).

Middleware (bench/src/middleware.bench.ts)

compileHandler chains of 1/3/5/10 handlers vs a plain array loop. This is the evidence behind design decision #5.

Parse (bench/src/parse.bench.ts)

parseQuery/parseCookies vs naive baselines.

WebSocket (bench/src/ws-bench.ts)

Pipelined echo clients (the ws library) hammering two servers — panmdaa and a bare ws WebSocketServer baseline (ws-servers.ts). Measures msg/s, avg/p99 latency, MB/s.

Current results

The [bench] claims in the docs are backed by these runs. To re-verify any claim:

npm run bench:router && npm run bench:http && npm run bench:ws

⚠️ Benchmarks are machine-relative. Re-run before quoting numbers in an issue or PR, and note the Node version and CPU used.