Coral vs AWS Compute Optimizer for EKS

Amazon EKS runs each managed node group as an EC2 Auto Scaling group. AWS Compute Optimizer makes rightsizing recommendations for Auto Scaling groups, but not for groups that run Amazon EKS workloads. It also makes no EC2 rightsizing recommendations for Spot Instances.

What an EKS node group can get is an idle recommendation. A group is idle only if each instance stayed within 5% peak CPU and 5 MB of network a day, over 14 days. Then Compute Optimizer suggests scaling the group down to one instance, or deleting it.

Why a node can look idle and still be full

Kubernetes places each pod by its CPU and memory requests, not by what it uses. The scheduler refuses a pod when the requests don’t fit, even if the node’s real use is very low.

So a node can show little CPU use and still have no room for another pod. Compute Optimizer reads the instance, not the requests. A type sized to the instance’s use can be too small for the requests its pods carry.

The saving starts with the requests. Size them to real use, and fewer EC2 instances can carry the same work. Kubernetes pod right sizing shows how, and Kubernetes cost optimization takes you from requests to nodes.

The two read different things, so you can run both. Here is where each one fits:

• Outside your clusters, use Compute Optimizer. It covers standalone EC2 instances, Lambda functions, RDS databases and more. You opt in, and it reads 14 days of CloudWatch metrics.
• Inside EKS, use Coral. It rebalances pod requests to real use, and empties spare nodes for Karpenter or the Cluster Autoscaler to remove. EKS cost optimization shows what connect sets up in your AWS account.
• To move a node group to a cheaper type yourself, create a new group, then drain the old nodes. How to change an EKS node group instance type has the steps. Our EKS pricing calculator shows what a node costs a month.
• For EBS data volumes on your VMs, Compute Optimizer recommends volume configurations. Amoeba moves a VM’s data volumes onto one btrfs pool sized to your data, then rightsizes it. See EBS cost optimization, and what Compute Optimizer recommends for EBS volumes.

For the rest of your AWS spend, see AWS cost optimization.

AWS Compute Optimizer User Guide, “What is AWS Compute Optimizer?” (opens in a new tab); AWS Compute Optimizer User Guide, “AWS resources supported by Compute Optimizer” (opens in a new tab); AWS Compute Optimizer User Guide, “Viewing idle resource recommendations” (opens in a new tab); AWS Compute Optimizer User Guide, “EC2 instance metrics” (opens in a new tab); AWS, “AWS Compute Optimizer FAQs” (opens in a new tab); AWS, “AWS Compute Optimizer Pricing” (opens in a new tab); Amazon EKS User Guide, “Simplify node lifecycle with managed node groups” (opens in a new tab); Kubernetes, “Resource Management for Pods and Containers” (opens in a new tab).