During development, you can throw money(=hardware) at the problem and get pretty decent results. In production, the startup time should not matter much, as warm

During development, you can throw money(=hardware) at the problem and get pretty decent results. In production, the startup time should not matter much, as warming of caches etc will take longer than tens of secs anyway. I guess the biggest issue is restarting services in production, but that is mostly alleviated by not having singular copies of crucial services.