AI summary by TwinMind
The fastest website ever?
Theo - t3․gg
30m 48s
McMaster Website Performance Analysis
- McMaster website goes viral every few months for being super fast
- Navigation feels like "flying" despite not using modern webtech
- Common misconception: fast because it uses vanilla HTML/JS instead of frameworks
- Reality: loads over 1.5MB of JavaScript on first page load
- Uses custom JavaScript layer for navigation, prefetching, and DOM manipulation
- Built on ASP.NET backend with custom client-side routing system
McMaster's Technical Implementation
- Prefetching Strategy: JavaScript tracks hover events, immediately prefetches content
- Response Format: Non-standard HTML with giant JSON objects at top of response
- Contains pile of numbers and data needed on client side
- Custom element naming following .NET conventions
- JavaScript Bundle: Over 1.5MB of minified JS code
- Custom navigation system for swapping page content
- Performance.mark() calls throughout codebase
- Two JS files fail to load but site still functions
- Page Speed Insights: 61% performance score, slow interaction-to-next-paint times
Next.js Faster Rewrite Project
- Performance: Comically faster than original McMaster site
- Bundle Size: Significantly less JavaScript than McMaster (300KB vs 1.5MB+)
- Contributors: Ethan Niser, Rees, Armen, plus Vercel team members
- Technology Stack: Next.js with Drizzle ORM, AI-generated mock data
Next.js Implementation Details
- Caching Strategy: unstable_cache with 2-hour revalidation periods
- Database Queries: Standard Drizzle queries with joins, nothing complex
- Static Generation: generate_static_params for build-time page generation
- Image Optimization: Next.js image component with quality=65, eager loading for first 15 images
Performance Optimizations
- Link Prefetching: prefetch=true on all navigation links
- Image Loading Strategy:
- First 15 images: loading="eager", decoding="sync"
- Remaining images: loading="lazy"
- Image count variable tracks render order
- Custom Link Component: 150-line file implementing advanced prefetching
- Intersection Observer for viewport-based prefetching
- Mouse enter triggers image preloading
- Mouse down navigation (faster than mouse up)
- Cancellable prefetch timeouts
Advanced Prefetching Features
- Viewport-based: Links prefetch when visible in viewport
- Hover Enhancement: Additional image preloading on mouse enter
- RSC Responses: JSON data fetched for route transitions
- Image Preloading: All images for next route loaded before click
- Network Throttling Test: Still fast on slow 4G compared to McMaster
Vercel Team Feedback & Future Plans
- Planned Upstream Features:
- CSS inlining for Tailwind (small CSS bundles)
- Mouse down navigation as Next.js option
- Cancellable router prefetch
- First-class image prefetching
- Default eager loading for first N images
- Accessibility Note: Mouse down navigation actually better for accessibility
- Reduces misclick probability for users with motor disabilities
- All accessibility tools trigger mouse down events
