Request functions vs always-on services
The core model difference. Vercel runs request-based functions that scale to zero; platforms like Railway, Render, Fly, and Heroku run always-on services (containers or instances) that never sleep. Workloads holding connections open, running between requests, or serving WebSockets long-term fit the second group — Vercel's own function duration and beta WebSocket limits state that boundary.
Evidence: [1] Vercel Functions limitations [2] Railway services [3] Render Web Services [4] Deploying with fly deploy