Development Startup Scaling
Method
pnpm benchmark:development-startup-scaling copies the repository's website/
configuration and plugin stack to a temporary site. It generates a deterministic
realistic vault at each size, including one homepage, and measures a first
process start followed by a stop and immediate second start.
The browser is started before timing. Each measurement includes the image pre-step and Vite server start, then records elapsed time from process spawn to:
- Vite accepting TCP connections
- the first document HTTP response
- visible
main#main-content
The benchmark does not delete existing user data or Vite caches. It removes only its temporary site and cache. Each value is one sample, not a median. Results are specific to this machine, configuration, and synthetic vault.
Results
Environment: Windows x64, Node.js v24.22.1, AMD Ryzen 7 7730U, 16 logical CPUs, 14,188 MB RAM.
All values are seconds from process spawn.
| Markdown files | First Vite ready | First HTTP | First visible | Second Vite ready | Second HTTP | Second visible |
|---|---|---|---|---|---|---|
| 10 | 22.11 | 37.46 | 37.53 | 21.98 | 35.24 | 35.25 |
| 100 | 22.23 | 42.93 | 42.96 | 21.23 | 35.99 | 36.01 |
| 1,000 | 21.49 | 62.85 | 62.88 | 21.24 | 37.47 | 37.48 |
| 2,000 | 21.65 | 85.99 | 86.02 | 21.45 | 39.64 | 39.65 |
| 4,000 | 22.38 | 135.32 | 135.35 | 21.80 | 43.41 | 43.43 |
| 5,000 | 21.84 | 156.30 | 156.33 | 21.47 | 43.88 | 43.89 |
| 10,000 | 22.06 | 271.34 | 271.37 | 23.31 | 52.81 | 52.82 |
Finding
The second start benefits substantially from the persistent development route
cache, while Vite startup remains nearly constant. At large vault sizes, cache
fingerprint reads must be bounded: unbounded reads caused EMFILE during the
second start at 5,000 files. The route-cache implementation therefore limits
fingerprint reads to the existing content-processing concurrency of 16.
Powered by Riebeckite