Companies overspend on Kubernetes thanks to underutilization of resources

Kubernetes is one of the most widely used container tools, but failure to accurately forecast resources leads to overprovision, waste and overspending.

A new report from Kubernetes automation platform CAST AI reveals high levels of underutilization are costing businesses more than necessary.

The analysis finds that for clusters containing 50 or more CPUs, organizations only utilize 13 percent of provisioned CPUs, and just 20 percent of memory, on average. The findings are based on CAST AI's analysis of 4,000 clusters running on Amazon Web Services (AWS), Google Cloud Platform (GCP), and Microsoft Azure (Azure) between January 1st and December 31st, 2023.

For larger clusters containing 1,000 or more CPUs, organizations on average are only using 17 percent of provisioned CPUs. CPU utilization varies little between AWS and Azure, they both share nearly identical utilization rates of 11 percent. Cloud waste is lower on Google, at 17 percent. For memory, differences are lower across providers: GCP (18 percent), AWS (20 percent), and Azure (22 percent).

The biggest drivers of overspending include, allocating more computing resources than necessary to an application or system, and requests for the number of CPUs being set too high. Many companies are reluctant to use Spot instances due to concerns over perceived instability. There's also low usage of 'custom instance size' on the Google Kubernetes Engine (GKE) making it difficult to choose the best CPU and memory ratio.

"This year's report makes it clear that companies running applications on Kubernetes are still in the early stages of their optimization journeys, and they're grappling with the complexity of manually managing cloud-native infrastructure," says Laurent Gil, co-founder and chief product officer at CAST AI. "The gap between provisioned and requested CPUs widened between 2022 and 2023 from 37 to 43 percent, so the problem is only going to worsen as more companies adopt Kubernetes."

The full report is available from the CAST AI site.

Image credit: aremafoto/depositphotos.com