Not sure, but I know that since the company I work for has switched to AWS, hosting costs have gone through the roof and performance has been reduced massively. Reason that I mention this is that the only clients that have actually benefited from the move are the larger, very high traffic clients, though even with those the difference is negligible for the huge increase in costs.
The more I think about it, the more I'm convinced that the vast majority of websites have nothing to gain from going the cloud way. Couldn't find any high traffic websites that are hosted on dedicated hardware to make a comparison though, hence the question.
vast majority of websites have nothing to gain from going the cloud way
The main benefit that I've seen from cloud computing is scalability, not price or performance. If you are running a website with predictable load, get a box that will handle that load plus some cushion. But if you have an app that has unpredictable load, the cloud will let you run 50 boxes one hour, and 2 the next. The savings come from not running 50 boxes when you only need 2.
It sounds like you are saying that you can "cloud burst". If anyone tells you that you can "autoscale", they're full of shit. It's true that you can "spin up" more instances on AWS, but that still takes time, even with the best automation.
The advantage you get from AWS is not related to the "cloud" (eg. virtual machines hosted elsewhere). The advantage is from the virtualization. If you have unpredictable load (like many people do) then you would actually benefit far more from having physical hardware utilizing virtualization.
If you think "scalability" is simply "spin up more AWS instances", you're wrong. In the time it takes for you to "cloud burst" from 2 instances to 50 instances (which, mind you, in most cases happens in an extremely short period of time... think measured in seconds), you're site will be screwed. Even if you have the best automation setup ever known, it still takes on the order of several minutes to get an instance of a machine up and running on AWS.
"So, the next time a vendor is like, 'It's going to be so cool, you can cloud burst your way to safety.' just tell him you are not a magical unicorn." - Adam Jacob (on Cloud Bursting and Other Magical Unicorns http://youtu.be/NTijokrOw6o )
Where I currently work (a stealth mode startup), we currently have a data aggregation engine running on between 20 and 120 instances. We take advantage of the ability to spin up instances quickly, not in response to a spike in traffic (since we don't have any yet), but in response to the size of the queue of jobs (which varies depending on what we've collected). We only need 120 running for an hour or so, then we terminate them. The cost savings we get is from not having to run 120 physical boxes all the time, and there is no downside if the queue grows a little while new instances spin up.
That seems to be a very small niche. Ex:
- shopping sites that have a significantly higher demand during holiday season
- tax filing type websites that see a huge demand during a certain day/week of the year
At my last job (ClearChannel), we had our own standalone datacenter in San Antonio with our own dedicated servers, as well as smaller server rooms in LA and NY. Our sites were all served by Akamai, so the dedicated servers were just origin servers.
