AWS Spot Instances are spare computing capacity that Amazon sells at a steep discount. They don’t make any money from idle servers, so Amazon is willing to sell excess cloud capacity at lower prices. Regardless of how you use EC2, there’s a way to lower your Amazon AWS bill with Spot Instances.
Depending on current computing demand on AWS, they can cost between 50% and 90% less than On-Demand. Using Amazon’s Advisor, you can estimate how much your business could save by relying less on Reserved or On-Demand Instances.
Experienced DevOp engineers can also determine which Instances are capable of handling your computing workload, but are less expensive. After hiring AWS Development experts, one AI company was able to reduce their cloud costs by 90% using older expensive Instance types instead.
More than you could use. In 2017, Clemson University set a record running 1,100,000 vCPUs on AWS Spot Instances at just one data center.
Western Digital recently ran a simulation that processed 2.5 million HPC tasks in just 8 hours.
Occasionally, Amazon will need to reclaim Spot Instances when there are too many requests for On-Demand Instances. If your bid isn’t high enough, the Instances will either be terminated, replaced or paused. The Instance may resume later, when enough spare capacity becomes available again.
On average, AWS Spot Instances are interrupted less than 5% of the time – but there can be significant variability depending on the Availability Zone and instance type.
Amazon’s Advisor provides specific estimates on how frequently you can expect service interruptions.
While it requires some additional DevOps to implement, if it weren’t for the occasional service interruption, there’d be no reason for AWS to offer discounts on Spot Instances.
Interruptions are managed using EC2 Spot Fleet. Engineers set the maximum bid your business is willing to pay to keep the Spot Instances running.
When the price exceeds your bid, the Instances will be paused, terminated or replaced depending on your business needs.
After harnessing the affordable power of Spot Instances, Yelp reduced the time it takes to run millions of tests from 2 days to 30 minutes. AdRoll scaled up their services while reducing operational costs by 83%. TLG Aerospace reduced the cost of each computational fluid dynamics simulation by 75%.
If you use EC2, then you can be cutting cloud costs by using Spot Instances.
They work well for containerized workloads or Kubernetes, CI/CD, Big Data, rendering pipelines, scaling applications and more.
Our team of AWS DevOps experts are happy to answer them: