bench-proto/¶

Early benchmark exploration using Tinybench. This component tests benchmarking libraries and approaches before the main React application is scaffolded.

Purpose¶

bench-proto/ serves as a sandbox for:

• Testing Tinybench as a potential benchmarking engine
• Exploring bun:sqlite for result storage
• Understanding browser benchmark methodology
• Informing library choices for the main React app

Current State¶

What's been explored¶

Tinybench API¶

import { Bench } from 'tinybench'

const bench = new Bench({ time: 1000 })

bench
  .add('faster task', () => { /* ... */ })
  .add('slower task', async () => { /* ... */ })

await bench.run()
console.table(bench.table())

Key findings¶

• Tinybench provides latency/throughput statistics (mean, median, p99, etc.)
• Supports async tasks and lifecycle hooks (beforeAll, afterAll)
• Lightweight (~10KB) with no dependencies
• Good integration with browser and Node.js environments

Planned improvements¶

• Test alternative benchmark libraries (benchmark.js, mitata)
• Integrate ML runtime adapters
• Add model load time measurement
• Implement result storage with bun:sqlite
• Create browser UI for running benchmarks

How to Run¶

cd bench-proto
bun install
bun run dev

Then open http://localhost:3456.

File structure¶

bench-proto/
├── index.html          Browser benchmark UI
├── server.ts           Bun static server
├── tinybench.js        Benchmark script example
├── PLAN.md             Implementation plan (now research/archive/plans/)
├── BRAINSTORM.md       Design discussions (now research/archive/plans/)
├── package.json        Dependencies
└── README.md           Documentation

Library evaluation criteria¶

When evaluating benchmark libraries for the main app:

Connection to thesis¶

The benchmark exploration in bench-proto/ informs:

• Chapter 3 (Methodology): Measurement procedure and iteration strategy
• Chapter 3a (Implementation): Metrics collection design
• Appendix: Benchmark result schema