Benchmarks
Apple Si dev machine, 10 physical cores — ratios and shapes, not absolutes. Every number traces to a named script + date (repo) — re-run it, don't trust it. Runtime-section numbers come from the experimental native runtime track, not the v0.1 default host.
Dev experience
| metric | value | context | run |
|---|---|---|---|
| dev cold start (hello-world) | 59ms | median; next dev: seconds; Vite-class | scripts/bench-dev-start.ts · 2026-06-10 |
| HMR save → flight pushed | 73ms | median 8/8, healthy at 400ms rapid-save cadence; browser apply +≈10ms; the push loop IS production live-RSC | scripts/bench-hmr.ts · 2026-06-10 |
Runtime
| metric | value | context | run |
|---|---|---|---|
| V8 isolate boot (snapshot) | 11.9ms | 10× vs no snapshot | PROBE=1 runtime-next/target/release/june-runtime · 2026-06-10 |
| bundled+snapshot first render | 21.8ms | the M3 target proof | PROBE=1 runtime-next/target/release/june-runtime · 2026-06-10 |
| cold start vs bare workerd | 82ms vs 78ms | spawn→first-200, same React-SSR hello fixture — par; warm trivial-page p50 3.1ms vs 0.35ms (pool-hop overhead = next lever) | bench/workerd-waterfall/measure.ts · 2026-06-10 |
| SSR load curve (POOL=8) | 4.7k rps peak | knee c≈16-32, p99 10ms at knee; ~470 rps/core whole-machine | bench/sweep.ts · 2026-06-10 |
Data layer
| metric | value | context | run |
|---|---|---|---|
| native sqlite point lookup (Rust op in V8) | 0.77µs | ≈ 1.1× bun:sqlite, 29× WASM | experiments/orm-bench (deno-sqlite-op) · 2026-06-10 |
| render-level auto-batch on Cloudflare D1 | 8.8× | vs concurrent per-component queries (11.5× vs sequential) | experiments/orm-bench/edge-d1 /render-bench · 2026-06-10 |
| Postgres pipelining | 48k ops/s | 3.9× sequential (12.2k), batch=50, single connection | experiments/orm-bench/pg-rust-pipeline · 2026-06-10 |