historical benchmarks for each pkc-js commit are saved to ./reports/<commit-hash>.txt and can be visualized at https://pkcprotocol.github.io/pkc-js-benchmarks/reports-ui

getting started

git clone https://github.com/pkcprotocol/pkc-js-benchmarks.git
npm install
npm start -- --runtime node --benchmark fetch-ipns

running specific benchmarks

npm start -- --runtime node|chrome --benchmark resolve-addresses|fetch-ipns|gateway-fetch-ipns|fetch-comment|publish|load-communities

load-communities benchmark

load-communities loads every production 5chan board over Helia/libp2p-js in pure-P2P browser mode and measures where the time goes (modeled on investigate_why_5chan_p2p_is_slow):

• the board list is discovered live from GitHub at runtime (nothing hardcoded) — see lib/discover-communities.ts
• boards are loaded in parallel (configurable concurrency) and it prints per-community load time plus a per-phase breakdown
• it snapshots peers & transports — how many connected peers a real browser can use (wss/webtransport/webrtc) and how many provider addresses are undialable from a browser

Runtime (node vs chrome) is one axis of the matrix; the config variants in benchmark-options.ts (e.g. concurrency) are the other. Run with no --runtime to execute every cell. node is the all-transports baseline (tcp/quic allowed — it loads over whatever transport works, like a non-browser peer). chrome is the real browser and can only use browser transports; there a connectionGater (countNonBrowserDials) counts how many provider addresses were undialable from the browser. So node-vs-chrome shows what a real browser loses.

npm start -- --benchmark load-communities                 # full matrix (node + chrome)
npm start -- --runtime chrome --benchmark load-communities

editing benchmark options (the pkc options used, the community addresses, etc)

edit the file ./benchmark-options.ts. this is needed to do manual debugging with specific pkc options.

print reports

reports are saved to ./report.json. running npm start overwrites the previous report.

npm run report
npm run report:inline

typecheck

npm run check

how it all works

• 1. npm start launches node ./start.ts (with optional arguments --runtime <runtime> --benchmark <benchmark>)
• 2. node ./start.ts launches:
  • 1. ./lib/server.ts which is needed to communicate with the browser benchmarks
  • 2. it reads ./benchmark-options.ts and iterates over all the benchmarks to do
  • 3. for each benchmark to do, it launches vitest in either node or a real browser (chromium via @vitest/browser-playwright) to execute it in an isolated environment (i.e. no pkc-js caching)
  • 4. it launches npm run report to print the last report (saved to ./report.json)

the TypeScript files are run directly by node (native type stripping, node ≥ 22.18) — there is no build step.