Elastic Computing

Sometimes there are new developments out there which concepts makes you tingle. They usually don't come twice from the same company. But thats the case with the S3 and EC2 from Amazon.

We already incoporated S3 into our applications. That makes us very easy on how to store binary data. But here comes EC2.

Grabbing a Debian image and modify it for out purpose and store it on S3 was just a bit of work. The image instance boots within 3 minutes into a full fledged application server with a simple command, and you can boot up as many as you like - figure that - an "unlimited" amount of servers at your fingertips to either hold to the load and serve the spikes. Fantastic. That makes me seriously happy. Forget about managing your machine park. You pay what you get. No need to maintain X machines for daily spikes or slashdotting events. No more idling heat emitters.

Some work has still to be done, of course. First, telling our load balancer that there are a new servers to serve. Second, observe the load of all servers running and deploy/shut-down servers on demand automatically. And if you are on Rails - guys from Thoughtworks wrote handy capistrano recipes interfacing with EC2

Some things have to be taken care of depending on what you want to do. First, data on these servers are transient - you shut it down, you loose the data you might have stored, not a problem in our case, but Amazon just started a limited beta with SimpleDB which addresses this problem. Second - it's still beta and we don't know how it performs real-life, but reports are great so far.

Last but not least there are many other applications for EC2. Think about on demand processing power for computing intensive applications. Computing cost start at 10 cent ($) the hour. That makes roughly $70 a month for a machine. Check the Amazon computing specs. Kind of a re-invention of renting mainframe time :-)

The future is bright - the future is ... elastic.