Ask YC: Scaling questions.

Ok, so, from most posts on the subject, it seems that people around these parts like AWS. Contrary to that opinion I chose (mt)(gs) to build my application on, this was partly due to funding, and the fact that mt is solidly $20/mo, every month.

The question: I'm growing, and I need to know how scaling _should_ work. I have a hosted, subscription based cms, there is one version of the toolset, hosted on the same server as a number of domains using the tools, I also provide e-mail via webmail and IMAP and POP (via (mt)).

As everybody knows, (mt) is a little slow, and a little prone to failure.

My current direction (needs work): When I start to max out one (gs) account, get a second one, run two copies of the program. (repeat as necessary)

This, to me will cause many headaches if I have more than say, two servers. (could be quite soon)

So, if my option were AWS, how would that change? I can't just add resources to an already running server, right? this just-in-time stuff is pretty new to me.

Would it be running the same way, but n servers could pull the app and client files from S3?

Could I have n number of servers pull an app, and files, and server x runs the app or starts a new server based on load?

How do the DNS issues work for AWS? What if my users want webmail, should that all be on a separate server instance, or can I scale that too?