Mike Raab at CopperEgg wrote about this situation last fall and it is worth considering.In the fairy tale of Goldilocks and the three bears, the heroine was trying to find the solution that was "just right." But when it comes to server provisioning in the cloud, do you often make the wrong choice, picking an instance that ends up being too costly for your needs?
Take a look at the current Linux pricing chart from AWS as an example. It shows numerous choices for your EC2 instance, ranging from a few pennies an hour to more than $2 per hour. If you run these VMs over the course of an entire month that works out to a range of $150 per month for el cheapo plan to more than $15 large. Granted, there are big differences in the instances that you are spinning up here: the micro two-cent plan doesn't include any storage and only has 600 MB of RAM, so chances are your database server isn't going to be running here.
At $150 a month, you probably can hide that in your expense report, and it is probably a lot less since you are only charged (at least on AWS) when your instance is actually running. But for the large loads, now we are talking some serious dough and it might make sense to examine your provisioning and see if you can get away with a lower pricing tier for your server.
There are lots of provisioning measurement tools out there, and perhaps now is a good time to see if you can find one that will help you cut your cloud costs. We have covered ones from CopperEgg, PuppetLabs, Nimbula and Opscode Chef, to name just a few of them.