Architecting for Scale, 2nd Edition

Part V. Tenet 5. Cloud: Utilizing the Cloud

Highly scaled, highly available applications require highly dynamic infrastructures.

As we build and architect highly available, highly scaled web applications, we also have to deal with highly variable load on our applications. From an infrastructure management standpoint, this has traditionally meant the need to overprovision infrastructure resources. If your application needs anywhere from 20 servers to 200 servers depending on the current numbers of users utilizing your application, then you better make sure you always have 200 servers available. In fact, you probably should have 250 servers available, in case your usage estimation was wrong. Failure to do so would mean you could suffer a scaling-related brownout or blackout, and your availability would suffer, and customers would be upset. How many times have you tried to use a website that was very popular at the moment only to find out that the website was unacceptably slow or unresponsive? This is the result of a scaling-related brownout or blackout, and it’s the consequence of incorrect resource planning for scale.

As the internet and our use of it matures, our use and expectations of sites go up. The ability to predict scaling needs for these applications becomes harder. Additionally, fewer companies can afford to have overprovisioned resources lying around unused during slow times.

The result? The need for highly dynamic infrastructures that can automatically size themselves ...