Amazon has now added a new layer to its utility computing platform with the Elastic Compute Cloud, EC2. S3 redefined the pricing model for utility storage, and EC2 looks set to do the same for utility and grid compute resources.
An array of virtual servers, EC2 allows you to run Linux images on its hardware. Currently Fedora Core 3 and Core 4 are supported, though Amazon expects any Linux distribution with a 2.6 kernel should run. Each image will give you the equivalent of a 1.7Ghz Xeon CPU with 1.75GB of RAM, 160GB of local storage, and 250Mb/s of connectivity.
The pricing model is attractive - especially when compared to Sun's $1 per CPU per hour. EC2 will cost $0.10 per instance-hour, and $0.20 per GB of externally transferred data.
While you still have to account for storage separately (as EC2 uses S3 as a storage platform), running a server flat out for a month with Amazon's new service will cost you around $72. However, if you need permanent infrastructure, buying your own hardware could well be more cost effective.
So why use EC2? If you need compute power quickly, and have an Amazon sanctioned image to hand, the EC2 is an ideal way of throwing a cluster of virtual servers at a particularly sticky problem that needs to be solved right now.
Alternatively, if you're prototyping a new web service and don't want to rent a full time server for your tests, a sacrificial server image on EC2 can save you time and money. You can keep EC2 costs to a minimum by treating it as a batch processing system, uploading and downloading batches of data to keep network charges down.
