Amazon EC2
I'm at the MIT Emerging Technology Conference this week. Jeff Bezos just got up and talked about some of the web services that Amazon is providing, such as Mechanical Turk, S3 and now EC2. This was the first time I really heard or understood what EC2 is and it is pretty mind-blowing. They are providing the ability to store an OS image in S3, then load it into EC2 to execute on demand. You get access to a 1.7 GHz x86 CPU, 1.7 GB RAM with 160 MB disk. Basically, they are providing a scalable server in the sky, where you pay per CPU hour and data transfer. For example, they are charging $0.10/hour for CPU, which is about $70/mo for 1 CPU, or, $70 for 700 CPU's for one hour.
Also, for those who haven't looked at it yet, S3 is also very cool. Basically a storage service with three calls: get, set, delete, where you provide a handle and a blob. Again, very scalable and replicated storage, where you pay as you go.
What excites me about all of this is that they are removing barriers to entry for all of us to get our really cool products out. We don't have to deal with what Jeff calls the muck, liking buying servers, settting up data centers and so on. As he put it, "We make muck, so you don't have to."
Check it out at aws.amazon.com.
Comments
I've been testing EC2, and it's quite good. The only flaw with the current model is that it doesn't offer persistent storage of data on the virtual hard drives. So while you can snapshot a disk image (which takes some time and effort, but works fine), you can't simply push the drive contents to a virtual image file when you shut down the virtual computer (nor when that virtual machine crashes).
I've suggested that Amazon think about both providing persistent availability or perhaps shared network-based storage for persistence; and that they add SQL database support with per-transaction pricing or perhaps CPU-based pricing.
Now I can load data at no cost from their S3 storage system--they don't charge for data passing between S3 and EC2. But that's dead storage.
With either persistent drive images, persistent network storage, or SQL databases that were persistent, I could dramatically expand some of my projects that are processor and system bound.
Posted by: Glenn Fleishman | September 27, 2006 09:14 AM