Scaling to a large number of users is a situation many plan for but rarely get to execute on. The simple answer for some is to throw more resources (and money) at the problem when the time does come.

I've been lucky enough to have a couple project go "Viral" resulting in millions of requests needing to be serviced. I'd love to share some of my experiences learning about:

1. What breaks down
2. What costs a lot
3. How you can prepare your models and environments for scaling up and down on a budget.