I'd like to help but you need to explain a bit more about what your system does. I've helped build and manage racks and cloud installs. Some general stuff:

Downtime happens. It often happens when you move hardware around (say, you need another SAN or replace you switches. Cloud servers are very ver flexible; you can set up a parallel system and test it for a week. You can't do that with your own hardware without spending serious cash. The first time I rearranged my entire server set up, while sitting in a cafe, was pretty damned cool.

It also happens when hardware fails. With owned systems you have to drive to the colo in the middle of the night.

I suggest you build a minimal parallel system in a cloud and try it out. Try several providers.

The Software is a web based PHP5 App running both Mysql and Oracle on top of SLES9. It collects performance information (KPIs) and sends out formatted reports (PDF/XLS) to clients & customers. It's been running on IBM Xseries 336 servers since 2005 (eek!) and really needs an upgrade. Our technical provider came up with the spec but am interested to see if cloud hosting has moved on; I understand it was a bit shakey to start with.

Zynga runs largely on EC2, something like 10,000 nodes. I don't know how to define "shaky", but that company bets hundreds of millions of dollars of revenue on EC2.

I didn't know that. I guess it scales extremely well.

Okay, you asked for it!

2 x Express x3550 M3 + 2 x Xeon 4C E5506 processors + 7 x 4Gb memory sticks + 2 x 73Gb disks + 1 x Qlogic raid card

1 x IBM storage DS3300 + 8 x 300Gb disks

2 x VM Ware standard 2 x SuSE 11

Also some Watchguard and Gigabit switch.

Formatting gone a bit wrong. So two servers with both cpu's, 7 sticks of 4gb ram and dual mirrored raid disks each.

Thanks for posting, I do love this community.

FWIW, I worked at an online advertising company that ran a whole division on EC2 machines - at the peak, nearly 100 instances. This was never a problem for us because we were judicious in our use of the machines (minify your content!) and we were disciplined in our deployments and management. If you go that route, just be sure the learn and follow best practices - you'll save yourself quite a lot of time.

Do both. You would be nuts to depend on a single cloud provider, as the model for clouds allows for adjusting "the dial" on performance to increase revenues.

Spend $30K on physical equipment and network, and $10K on cloud deployment.

Then, over the course of a year or two, as you expand and learn how the cloud "really works", you can adjust as needed.

It very much depends on the nature of the usage, storage and bandwidth are the big killers for some cloud applications.

For me (live video) the cloud doesn't even get close to being competitive with leased hardware.

1GBit flat rate with 20T of storage goes for about 1200 euros / months, including a quad (or even eight) core machine.

Well ... isn't the question more "Will $40K of IBM and VMWare kit in our hands be (significantly) more reliable than Amazon's services?"

For your purposes, including scheduled downtime.

Also factor into the cost of the time it will take your people to learn how to make your systems on EC2 reliable enough.

I think that's an interesting point. However, I do have to consider running sensitive performance data running on equipment we don't control or host. So, is it wise to run customers data on kit we don't own or host?