Automated scaling comes to Amazon EC2

Amazon have, in our opinion, always offered the best solution for cloud-deployment of services, including web-sites, and one of the primary reasons was the speed with which you can start, stop and upgrade your capacity.

However, Amazon have again raised their game and added automated scaling. This is neat and gives you two approaches, by schedule and by demand. Both work in very similar ways with scheduled scaling allowing you to increase and decrease the number of instances on a predictable schedule. Typically that might mean lowering the grunt overnight and raising it at peak times such as 9am or the 5pm to 8pm tea-time rush. If you’ve got a quiet overnight server load you could easily reduce your operating costs by 25% or more.

Method two is scaling based on demand. You set specific policies based on many factors (including CPU usage, bandwidth, memory) to start and stop instances. It only take a minute to start new instances (what Amazon call servers) so with only a momentary slow down you could easily build a skeleton server farm that fires up 100% on demand.

In practical terms you can now easily plan your cloud capacity to meet your digital needs. Got a huge email drop going out? Up the capacity. Do you get little overnight traffic? Shutdown most of your services. An unexpected spike in traffic caused by an item going viral? No problem, the server capacity will scale up and then back down again once the flurry has passed.

It seems such an obvious thing to do but until now you’ve broadly had to build your digital capacity based on peak usage, and then leave a large part of it unused during the off-peak periods. It’s a lot like the old days of fitting out and maintaining a 50-man office just because you have 50 staffers who converge at the same time for a few hours a week. Nowadays, of course, we implement hot desking and location independent working so we only need the bare minimum of office space and facilities.

We estimate that since we started using Amazon services, the cost per event (how we measure the cost of server power) has come down in real terms by 75%.

That’s a three fold reduction in cost in just over 3 years. We share these saving with our clients, so our typical server configuration (couple of servers, working in tandem) has come down in cost from circa £1,200 per month to less than £300 and a typical small installation server cost has come down from £100 a month to £30.

Screen Shot 2013-12-11 at 11.10.29