The effort is really not that much - it's like 20 minutes per server, and each one can service a lot of users. And since the application is broken up into sever

The effort is really not that much - it's like 20 minutes per server, and each one can service a lot of users. And since the application is broken up into several independent components, they can be scaled separately according to their particular demands. The needs change depending on where the bottlenecks are, so it's not as simple as X users = Y capacity, but this is more helpful than not - because it lets me use more fine-grained scale planning.