If Amazon has two hundred r4 servers at its Data Center in Ohio, but their customers are only using one hundred and fifty right now – what does it do with the rest of them? The answer is simple: AWS sells the remaining computing capacity on the fifty servers at a steep discount, hoping customers will now use them. We call this idle computing capacity, sold at a steep discount, AWS Spot Instances.
How much cheaper are Spot Instances? Well it depends a lot on what kind of computing power you need from AWS, but up to 90% less expensive. You can get a better idea of how much cheaper Spot Instances are when compared to On-Demand or Reserved Instances, by checking Amazon’s Spot Instance Advisor here.
For example, if your app or Kubernetes cluster could run just fine on r4.xlarge servers in Ohio – then you could be using Spot Instances to save 85% on your cloud costs when compared to On-Demand prices. Or if your Big Data project needed the high memory computing power of r5ad.24xlarge servers in Central Canada, you could save 78% when compared to On-Demand prices. The pricing can fluctuate, so make sure to do your research before estimating budgets for your cloud services.
No matter the EC2 instance you use, there is almost certainly a way to take advantage of Spot Instances to lower your Amazon AWS billing. In fact, many businesses are able to grow their app while shrinking their cloud costs with Spot Instances. If you normally pay .11 cents per hour for an On-Demand instance, it’s possible you could only pay .03 to .04 cents which can save significant amounts of money over the course of a month or year.
So why doesn’t everyone just use Spot Instances instead of On-Demand or Reserved? Because Amazon can kick you off the instances, with only 2 minutes warning, when demand for On-Demand and Reserved gets too high. AWS isn’t going to offer steeply discounted computing power without there being a drawback…otherwise no one would pay “full” price to use their servers.
Can your app, website or project survive an interruption to its computing power? For batch work or anything fault-tolerant the answer could simply be yes. But for many projects the answer is a lot more complicated. That’s when you need to EC2 Spot Fleet: it automatically manages how your cloud services respond when you’re Spot Instances get taken back by Amazon.
The cost savings of Spot Instances over time can be well worth it, but first you’ll need to invest in some engineers who are familiar with AWS DevOps. These engineers can help configure your Spot Fleet to find alternative instances or availability zones that replace your lost Spot Instances. You can see the average frequency of interruption for an instance type and its availability zone by checking the Spot Instance Advisor, but most types are interrupted less than 5% of the time.
Essentially you, and all of AWS other customers using Spot Instances, are bidding on the maximum amount price you’re willing to pay for the computing power – up until it becomes just as expensive as an On-Demand Instance. The demand on Amazon’s servers is constantly in flux, but there will always be computing capacity for those willing to outbid their competitors – but the right DevOps engineers know how to get the computing power from the right types of diverse instances to run your project, without needing to pay full price.
Huge enterprises like Zillow, Yelp, and Lyft have successfully transformed their businesses by relying less on Reserved Instances and more on Spot Instances – but you don’t need to be that big to benefit from lower cloud costs or more computing power. Just because NASA has done it, doesn’t mean you need to be a rocket scientist to save money: you just need to find the right team of AWS experts to identify what type of instances your project needs, how much you’re willing to pay for them and configure it all in EC2 Spot Fleet.
So regardless if your project or app is short on computing power for batchwork or not – thousands of businesses have figured out how to save money on cloud costs using Spot Instances – despite their service interruptions.
What could your business do if it was spending 50 to 90% less on it’s cloud costs?