Data-driven Amazon EKS cost optimization: A practical guide to workload analysis

This article provides a practical guide to optimizing Amazon EKS costs through data-driven workload analysis, identifying three primary sources of resource waste.

• Greedy workloads: Oversized pod resource requests waste capacity; rightsizing to actual usage can reduce instances by two-thirds
• Pet workloads: Overly strict topology spread constraints and pod disruption budgets prevent node consolidation; relaxing maxSkew from 1 to 3 reduces replicas by 43%
• Isolated workloads: Fragmented NodePools create capacity silos; consolidating to single NodePools reduces total capacity needs by 22%
• Use tools like Kubecost, Goldilocks, and Vertical Pod Autoscaler for rightsizing recommendations
• Define constraints at pod level rather than creating multiple NodePools to maintain flexibility
• Audit PDB configurations and do-not-disrupt annotations to enable cost-saving operations

The guide emphasizes eliminating waste through data-driven optimization while maintaining performance and reliability across production EKS environments.