StackOverdrive’s clients commonly save 30%-40% on infrastructure costs after StackOverdrive completes the client’s infrastructure migration or cost engineers the client’s infrastructure. Although actual savings depend on the state of your infrastructure beforehand, StackOverdrive designs for robustness, scalability, and security, and in doing so we are usually able to create infrastructure that is significantly cheaper than common naive infrastructure patterns.
The most common way that organizations provision cloud computing resources is by provisioning a specific, fixed number of computer instances with their cloud providers, using the on demand option (or equivalent). To achieve easier management, typically each instance will be provisioned for a single purpose, and the number of instances will be chosen to hit a theoretical or historical maximum load.
Let’s talk about the problems with this kind of setup, and how we solve them.
Knowing where your money is going is a crucial component of controlling costs. AWS provides good reporting, but they don’t know what you use your infrastructure for. One key best practice that StackOverdrive always implements is to tag resources. That way, you can pivot on those tags to understand exactly how the spending is broken down.
On demand instances in AWS (and other providers’ equivalents) are the most expensive way to buy computing resources. They are available instantly, guaranteed not to go away without warning, and drain your pocket.
StackOverdrive typically uses redundant spot instances across availability zones to provide you computers at a fraction of the cost per instance – typically 10% of the cost.
If you’re provisioning for peak usage, by definition you’re wasting money at all other times. Even worse, when your marketing department pulls off a coup and they don’t tell engineering, your systems are vulnerable to being overwhelmed. By implementing auto scaling we ensure that you’re always using the right number of resources (and if your traffic blows up, you’ll be ready).
Some other benefits of autos caling is that it lets you provision spot instances across availability zones, improving your cost and your reliability.
For all kinds of good reasons, if you’re dealing with fixed computers, you’re keeping one application type per computer. But that’s still costing you money, because you’re using too many computers (and the expensive on demand kind).
StackOverdrive strongly recommends containerizing applications, and using computer clusters to run those containers. This way, you only consume the number of computers you need to run all your applications. This also feeds into working with spot instances and auto scaling – your cluster scales to overall activity, and it is designed to handle spot instances going away.
We are happy to review your infrastructure with a view to bringing down costs with or without doing a full migration to a new infrastructure. The above are some common anti-patterns we encounter and simple ways we can help you control your cloud costs. With a full review of your infrastructure we can find the problems and solutions in your specific setup.